Add Depth Anything (#28654)

* First draft

* More improvements

* More improvements

* More improvements

* More improvements

* Add docs

* Remove file

* Add copied from

* Address comments

* Address comments

* Address comments

* Fix style

* Update docs

* Convert all checkpoints, add integration test

* Rename checkpoints

* Add pretrained backbone attributes

* Fix default config

* Address comment

* Add figure to docs

* Fix bug thanks to @xenova

* Update conversion script

* Fix integration test

README.md

@@ -339,6 +339,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (from University of Hong Kong and TikTok) released with the paper [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) by Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao.
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.

README_es.md

@@ -314,6 +314,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (from University of Hong Kong and TikTok) released with the paper [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) by Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao.
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.

README_hd.md

@@ -288,6 +288,7 @@ conda install conda-forge::transformers
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (सेंसटाइम रिसर्च से) साथ में पेपर [डिफॉर्मेबल डीईटीआर: डिफॉर्मेबल ट्रांसफॉर्मर्स फॉर एंड-टू-एंड ऑब्जेक्ट डिटेक्शन] (https://arxiv.org/abs/2010.04159) Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, जिफेंग दाई द्वारा पोस्ट किया गया।
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (फेसबुक से) साथ में पेपर [ट्रेनिंग डेटा-एफिशिएंट इमेज ट्रांसफॉर्मर और डिस्टिलेशन थ्रू अटेंशन](https://arxiv .org/abs/2012.12877) ह्यूगो टौव्रोन, मैथ्यू कॉर्ड, मैथिज्स डूज़, फ़्रांसिस्को मस्सा, एलेक्ज़ेंडर सबलेरोल्स, हर्वे जेगौ द्वारा।
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (Google AI से) Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun. द्वाराअनुसंधान पत्र [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) के साथ जारी किया गया
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (University of Hong Kong and TikTok से) Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao. द्वाराअनुसंधान पत्र [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) के साथ जारी किया गया
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (फेसबुक से) साथ में कागज [ट्रांसफॉर्मर्स के साथ एंड-टू-एंड ऑब्जेक्ट डिटेक्शन](https://arxiv. org/abs/2005.12872) निकोलस कैरियन, फ़्रांसिस्को मस्सा, गेब्रियल सिनेव, निकोलस उसुनियर, अलेक्जेंडर किरिलोव, सर्गेई ज़ागोरुयको द्वारा।
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (माइक्रोसॉफ्ट रिसर्च से) कागज के साथ [DialoGPT: बड़े पैमाने पर जनरेटिव प्री-ट्रेनिंग फॉर कन्वर्सेशनल रिस्पांस जेनरेशन](https ://arxiv.org/abs/1911.00536) यिज़े झांग, सिकी सन, मिशेल गैली, येन-चुन चेन, क्रिस ब्रोकेट, जियांग गाओ, जियानफेंग गाओ, जिंगजिंग लियू, बिल डोलन द्वारा।

README_ja.md

@@ -348,6 +348,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (SenseTime Research から) Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai から公開された研究論文: [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159)
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (Facebook から) Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou から公開された研究論文: [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877)
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (Google AI から) Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun. から公開された研究論文 [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505)
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (University of Hong Kong and TikTok から) Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao. から公開された研究論文 [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891)
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (The University of Texas at Austin から) Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl. から公開された研究論文 [NMS Strikes Back](https://arxiv.org/abs/2212.06137)
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (Facebook から) Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko から公開された研究論文: [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872)
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (Microsoft Research から) Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan から公開された研究論文: [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536)

README_ko.md

@@ -263,6 +263,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (SenseTime Research 에서) Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai 의 [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) 논문과 함께 발표했습니다.
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (Facebook 에서) Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou 의 [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) 논문과 함께 발표했습니다.
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (Google AI 에서 제공)은 Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.의 [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505)논문과 함께 발표했습니다.
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (University of Hong Kong and TikTok 에서 제공)은 Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao.의 [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891)논문과 함께 발표했습니다.
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (The University of Texas at Austin 에서 제공)은 Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.의 [NMS Strikes Back](https://arxiv.org/abs/2212.06137)논문과 함께 발표했습니다.
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (Facebook 에서) Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko 의 [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) 논문과 함께 발표했습니다.
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (Microsoft Research 에서) Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan 의 [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) 논문과 함께 발표했습니다.

README_zh-hans.md

@@ -287,6 +287,7 @@ conda install conda-forge::transformers
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (来自 SenseTime Research) 伴随论文 [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) 由 Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai 发布。
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (来自 Facebook) 伴随论文 [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) 由 Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou 发布。
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (来自 Google AI) 伴随论文 [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) 由 Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun 发布。
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (来自 University of Hong Kong and TikTok) 伴随论文 [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) 由 Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao 发布。
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (来自 The University of Texas at Austin) 伴随论文 [NMS Strikes Back](https://arxiv.org/abs/2212.06137) 由 Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl 发布。
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (来自 Facebook) 伴随论文 [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) 由 Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko 发布。
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (来自 Microsoft Research) 伴随论文 [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) 由 Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan 发布。

README_zh-hant.md

@@ -299,6 +299,7 @@ conda install conda-forge::transformers
 1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
 1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
 1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
+1. **[Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)** (from University of Hong Kong and TikTok) released with the paper [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) by Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao.
 1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
 1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
 1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.

docs/source/en/_toctree.yml

@@ -525,6 +525,8 @@
         title: Deformable DETR
       - local: model_doc/deit
         title: DeiT
+      - local: model_doc/depth_anything
+        title: Depth Anything
       - local: model_doc/deta
         title: DETA
       - local: model_doc/detr

docs/source/en/index.md

@@ -112,6 +112,7 @@ Flax), PyTorch, and/or TensorFlow.
 |               [Deformable DETR](model_doc/deformable_detr)               |       ✅        |         ❌         |      ❌      |
 |                          [DeiT](model_doc/deit)                          |       ✅        |         ✅         |      ❌      |
 |                        [DePlot](model_doc/deplot)                        |       ✅        |         ❌         |      ❌      |
+|                [Depth Anything](model_doc/depth_anything)                |       ✅        |         ❌         |      ❌      |
 |                          [DETA](model_doc/deta)                          |       ✅        |         ❌         |      ❌      |
 |                          [DETR](model_doc/detr)                          |       ✅        |         ❌         |      ❌      |
 |                      [DialoGPT](model_doc/dialogpt)                      |       ✅        |         ✅         |      ✅      |

docs/source/en/model_doc/depth_anything.md

+<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Depth Anything
+
+## Overview
+
+The Depth Anything model was proposed in [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) by Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao. Depth Anything is based on the [DPT](dpt) architecture, trained on ~62 million images, obtaining state-of-the-art results for both relative and absolute depth estimation.
+
+The abstract from the paper is the following:
+
+*This work presents Depth Anything, a highly practical solution for robust monocular depth estimation. Without pursuing novel technical modules, we aim to build a simple yet powerful foundation model dealing with any images under any circumstances. To this end, we scale up the dataset by designing a data engine to collect and automatically annotate large-scale unlabeled data (~62M), which significantly enlarges the data coverage and thus is able to reduce the generalization error. We investigate two simple yet effective strategies that make data scaling-up promising. First, a more challenging optimization target is created by leveraging data augmentation tools. It compels the model to actively seek extra visual knowledge and acquire robust representations. Second, an auxiliary supervision is developed to enforce the model to inherit rich semantic priors from pre-trained encoders. We evaluate its zero-shot capabilities extensively, including six public datasets and randomly captured photos. It demonstrates impressive generalization ability. Further, through fine-tuning it with metric depth information from NYUv2 and KITTI, new SOTAs are set. Our better depth model also results in a better depth-conditioned ControlNet.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/depth_anything_overview.jpg"
+alt="drawing" width="600"/>
+
+<small> Depth Anything overview. Taken from the <a href="https://arxiv.org/abs/2401.10891">original paper</a>.</small>
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr).
+The original code can be found [here](https://github.com/LiheYoung/Depth-Anything).
+
+## Usage example
+
+There are 2 main ways to use Depth Anything: either using the pipeline API, which abstracts away all the complexity for you, or by using the `DepthAnythingForDepthEstimation` class yourself.
+
+### Pipeline API
+
+The pipeline allows to use the model in a few lines of code:
+
+```python
+>>> from transformers import pipeline
+>>> from PIL import Image
+>>> import requests
+
+>>> # load pipe
+>>> pipe = pipeline(task="depth-estimation", model="LiheYoung/depth-anything-small-hf")
+
+>>> # load image
+>>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
+>>> image = Image.open(requests.get(url, stream=True).raw)
+
+>>> # inference
+>>> depth = pipe(image)["depth"]
+```
+
+### Using the model yourself
+
+If you want to do the pre- and postprocessing yourself, here's how to do that:
+
+```python
+>>> from transformers import AutoImageProcessor, AutoModelForDepthEstimation
+>>> import torch
+>>> import numpy as np
+>>> from PIL import Image
+>>> import requests
+
+>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+>>> image = Image.open(requests.get(url, stream=True).raw)
+
+>>> image_processor = AutoImageProcessor.from_pretrained("LiheYoung/depth-anything-small-hf")
+>>> model = AutoModelForDepthEstimation.from_pretrained("LiheYoung/depth-anything-small-hf")
+
+>>> # prepare image for the model
+>>> inputs = image_processor(images=image, return_tensors="pt")
+
+>>> with torch.no_grad():
+...     outputs = model(**inputs)
+...     predicted_depth = outputs.predicted_depth
+
+>>> # interpolate to original size
+>>> prediction = torch.nn.functional.interpolate(
+...     predicted_depth.unsqueeze(1),
+...     size=image.size[::-1],
+...     mode="bicubic",
+...     align_corners=False,
+... )
+
+>>> # visualize the prediction
+>>> output = prediction.squeeze().cpu().numpy()
+>>> formatted = (output * 255 / np.max(output)).astype("uint8")
+>>> depth = Image.fromarray(formatted)
+```
+
+## DepthAnythingConfig
+
+[[autodoc]] DepthAnythingConfig
+
+## DepthAnythingForDepthEstimation
+
+[[autodoc]] DepthAnythingForDepthEstimation
+    - forward
\ No newline at end of file

docs/source/en/tasks/monocular_depth_estimation.md

@@ -30,7 +30,7 @@ The task illustrated in this tutorial is supported by the following model archit
 
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
 
-[DPT](../model_doc/dpt), [GLPN](../model_doc/glpn)
+[Depth Anything](../model_doc/depth_anything), [DPT](../model_doc/dpt), [GLPN](../model_doc/glpn)
 
 <!--End of the generated tip-->
 

src/transformers/__init__.py

@@ -373,6 +373,7 @@ _import_structure = {
         "TransfoXLTokenizer",
     ],
     "models.deprecated.van": ["VAN_PRETRAINED_CONFIG_ARCHIVE_MAP", "VanConfig"],
+    "models.depth_anything": ["DEPTH_ANYTHING_PRETRAINED_CONFIG_ARCHIVE_MAP", "DepthAnythingConfig"],
     "models.deta": ["DETA_PRETRAINED_CONFIG_ARCHIVE_MAP", "DetaConfig"],
     "models.detr": ["DETR_PRETRAINED_CONFIG_ARCHIVE_MAP", "DetrConfig"],
     "models.dialogpt": [],
@@ -1985,6 +1986,13 @@ else:
             "VanPreTrainedModel",
         ]
     )
+    _import_structure["models.depth_anything"].extend(
+        [
+            "DEPTH_ANYTHING_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "DepthAnythingForDepthEstimation",
+            "DepthAnythingPreTrainedModel",
+        ]
+    )
     _import_structure["models.deta"].extend(
         [
             "DETA_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -5097,6 +5105,7 @@ if TYPE_CHECKING:
         TransfoXLTokenizer,
     )
     from .models.deprecated.van import VAN_PRETRAINED_CONFIG_ARCHIVE_MAP, VanConfig
+    from .models.depth_anything import DEPTH_ANYTHING_PRETRAINED_CONFIG_ARCHIVE_MAP, DepthAnythingConfig
     from .models.deta import DETA_PRETRAINED_CONFIG_ARCHIVE_MAP, DetaConfig
     from .models.detr import DETR_PRETRAINED_CONFIG_ARCHIVE_MAP, DetrConfig
     from .models.dinat import DINAT_PRETRAINED_CONFIG_ARCHIVE_MAP, DinatConfig
@@ -6600,6 +6609,11 @@ if TYPE_CHECKING:
             VanModel,
             VanPreTrainedModel,
         )
+        from .models.depth_anything import (
+            DEPTH_ANYTHING_PRETRAINED_MODEL_ARCHIVE_LIST,
+            DepthAnythingForDepthEstimation,
+            DepthAnythingPreTrainedModel,
+        )
         from .models.deta import (
             DETA_PRETRAINED_MODEL_ARCHIVE_LIST,
             DetaForObjectDetection,

src/transformers/models/__init__.py

@@ -64,6 +64,7 @@ from . import (
     deformable_detr,
     deit,
     deprecated,
+    depth_anything,
     deta,
     detr,
     dialogpt,

src/transformers/models/auto/configuration_auto.py

@@ -76,6 +76,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("decision_transformer", "DecisionTransformerConfig"),
         ("deformable_detr", "DeformableDetrConfig"),
         ("deit", "DeiTConfig"),
+        ("depth_anything", "DepthAnythingConfig"),
         ("deta", "DetaConfig"),
         ("detr", "DetrConfig"),
         ("dinat", "DinatConfig"),
@@ -308,6 +309,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("deberta-v2", "DEBERTA_V2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("deformable_detr", "DEFORMABLE_DETR_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("deit", "DEIT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("depth_anything", "DEPTH_ANYTHING_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("deta", "DETA_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("detr", "DETR_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("dinat", "DINAT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -532,6 +534,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("deformable_detr", "Deformable DETR"),
         ("deit", "DeiT"),
         ("deplot", "DePlot"),
+        ("depth_anything", "Depth Anything"),
         ("deta", "DETA"),
         ("detr", "DETR"),
         ("dialogpt", "DialoGPT"),

src/transformers/models/auto/image_processing_auto.py

@@ -54,6 +54,7 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("data2vec-vision", "BeitImageProcessor"),
         ("deformable_detr", "DeformableDetrImageProcessor"),
         ("deit", "DeiTImageProcessor"),
+        ("depth_anything", "DPTImageProcessor"),
         ("deta", "DetaImageProcessor"),
         ("detr", "DetrImageProcessor"),
         ("dinat", "ViTImageProcessor"),

src/transformers/models/auto/modeling_auto.py

@@ -684,6 +684,7 @@ MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING_NAMES = OrderedDict(
 MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES = OrderedDict(
     [
         # Model for depth estimation mapping
+        ("depth_anything", "DepthAnythingForDepthEstimation"),
         ("dpt", "DPTForDepthEstimation"),
         ("glpn", "GLPNForDepthEstimation"),
     ]

src/transformers/models/depth_anything/__init__.py

+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import TYPE_CHECKING
+
+from ...file_utils import _LazyModule, is_torch_available
+from ...utils import OptionalDependencyNotAvailable
+
+
+_import_structure = {
+    "configuration_depth_anything": ["DEPTH_ANYTHING_PRETRAINED_CONFIG_ARCHIVE_MAP", "DepthAnythingConfig"]
+}
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_depth_anything"] = [
+        "DEPTH_ANYTHING_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "DepthAnythingForDepthEstimation",
+        "DepthAnythingPreTrainedModel",
+    ]
+
+
+if TYPE_CHECKING:
+    from .configuration_depth_anything import DEPTH_ANYTHING_PRETRAINED_CONFIG_ARCHIVE_MAP, DepthAnythingConfig
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_depth_anything import (
+            DEPTH_ANYTHING_PRETRAINED_MODEL_ARCHIVE_LIST,
+            DepthAnythingForDepthEstimation,
+            DepthAnythingPreTrainedModel,
+        )
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)

src/transformers/models/depth_anything/configuration_depth_anything.py

+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" DepthAnything model configuration"""
+
+import copy
+
+from ...configuration_utils import PretrainedConfig
+from ...utils import logging
+from ..auto.configuration_auto import CONFIG_MAPPING
+
+
+logger = logging.get_logger(__name__)
+
+DEPTH_ANYTHING_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "LiheYoung/depth-anything-small-hf": "https://huggingface.co/LiheYoung/depth-anything-small-hf/resolve/main/config.json",
+}
+
+
+class DepthAnythingConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`DepthAnythingModel`]. It is used to instantiate an DepthAnything
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the DepthAnything
+    [LiheYoung/depth-anything-small-hf](https://huggingface.co/LiheYoung/depth-anything-small-hf) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        backbone_config (`Union[Dict[str, Any], PretrainedConfig]`, *optional*):
+            The configuration of the backbone model. Only used in case `is_hybrid` is `True` or in case you want to
+            leverage the [`AutoBackbone`] API.
+        backbone (`str`, *optional*):
+            Name of backbone to use when `backbone_config` is `None`. If `use_pretrained_backbone` is `True`, this
+            will load the corresponding pretrained weights from the timm or transformers library. If `use_pretrained_backbone`
+            is `False`, this loads the backbone's config and uses that to initialize the backbone with random weights.
+        use_pretrained_backbone (`bool`, *optional*, defaults to `False`):
+            Whether to use pretrained weights for the backbone.
+        patch_size (`int`, *optional*, defaults to 14):
+            The size of the patches to extract from the backbone features.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        reassemble_hidden_size (`int`, *optional*, defaults to 384):
+            The number of input channels of the reassemble layers.
+        reassemble_factors (`List[int]`, *optional*, defaults to `[4, 2, 1, 0.5]`):
+            The up/downsampling factors of the reassemble layers.
+        neck_hidden_sizes (`List[str]`, *optional*, defaults to `[48, 96, 192, 384]`):
+            The hidden sizes to project to for the feature maps of the backbone.
+        fusion_hidden_size (`int`, *optional*, defaults to 64):
+            The number of channels before fusion.
+        head_in_index (`int`, *optional*, defaults to -1):
+            The index of the features to use in the depth estimation head.
+        head_hidden_size (`int`, *optional*, defaults to 32):
+            The number of output channels in the second convolution of the depth estimation head.
+
+    Example:
+
+    ```python
+    >>> from transformers import DepthAnythingConfig, DepthAnythingForDepthEstimation
+
+    >>> # Initializing a DepthAnything small style configuration
+    >>> configuration = DepthAnythingConfig()
+
+    >>> # Initializing a model from the DepthAnything small style configuration
+    >>> model = DepthAnythingForDepthEstimation(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "depth_anything"
+
+    def __init__(
+        self,
+        backbone_config=None,
+        backbone=None,
+        use_pretrained_backbone=False,
+        patch_size=14,
+        initializer_range=0.02,
+        reassemble_hidden_size=384,
+        reassemble_factors=[4, 2, 1, 0.5],
+        neck_hidden_sizes=[48, 96, 192, 384],
+        fusion_hidden_size=64,
+        head_in_index=-1,
+        head_hidden_size=32,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        if use_pretrained_backbone:
+            raise ValueError("Pretrained backbones are not supported yet.")
+
+        if backbone_config is not None and backbone is not None:
+            raise ValueError("You can't specify both `backbone` and `backbone_config`.")
+
+        if backbone_config is None and backbone is None:
+            logger.info("`backbone_config` is `None`. Initializing the config with the default `Dinov2` backbone.")
+            backbone_config = CONFIG_MAPPING["dinov2"](
+                image_size=518,
+                hidden_size=384,
+                num_attention_heads=6,
+                out_indices=[9, 10, 11, 12],
+                apply_layernorm=True,
+                reshape_hidden_states=False,
+            )
+        elif isinstance(backbone_config, dict):
+            backbone_model_type = backbone_config.get("model_type")
+            config_class = CONFIG_MAPPING[backbone_model_type]
+            backbone_config = config_class.from_dict(backbone_config)
+
+        self.backbone_config = backbone_config
+        self.backbone = backbone
+        self.use_pretrained_backbone = use_pretrained_backbone
+        self.reassemble_hidden_size = reassemble_hidden_size
+        self.patch_size = patch_size
+        self.initializer_range = initializer_range
+        self.reassemble_factors = reassemble_factors
+        self.neck_hidden_sizes = neck_hidden_sizes
+        self.fusion_hidden_size = fusion_hidden_size
+        self.head_in_index = head_in_index
+        self.head_hidden_size = head_hidden_size
+
+    def to_dict(self):
+        """
+        Serializes this instance to a Python dictionary. Override the default [`~PretrainedConfig.to_dict`]. Returns:
+            `Dict[str, any]`: Dictionary of all the attributes that make up this configuration instance,
+        """
+        output = copy.deepcopy(self.__dict__)
+
+        if output["backbone_config"] is not None:
+            output["backbone_config"] = self.backbone_config.to_dict()
+
+        output["model_type"] = self.__class__.model_type
+        return output

src/transformers/models/depth_anything/convert_depth_anything_to_hf.py

+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Convert Depth Anything checkpoints from the original repository. URL:
+https://github.com/LiheYoung/Depth-Anything"""
+
+
+import argparse
+from pathlib import Path
+
+import requests
+import torch
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from transformers import DepthAnythingConfig, DepthAnythingForDepthEstimation, Dinov2Config, DPTImageProcessor
+from transformers.utils import logging
+
+
+logging.set_verbosity_info()
+logger = logging.get_logger(__name__)
+
+
+def get_dpt_config(model_name):
+    if "small" in model_name:
+        backbone_config = Dinov2Config.from_pretrained(
+            "facebook/dinov2-small", out_indices=[9, 10, 11, 12], apply_layernorm=True, reshape_hidden_states=False
+        )
+        fusion_hidden_size = 64
+        neck_hidden_sizes = [48, 96, 192, 384]
+    elif "base" in model_name:
+        backbone_config = Dinov2Config.from_pretrained(
+            "facebook/dinov2-base", out_indices=[9, 10, 11, 12], apply_layernorm=True, reshape_hidden_states=False
+        )
+        fusion_hidden_size = 128
+        neck_hidden_sizes = [96, 192, 384, 768]
+    elif "large" in model_name:
+        backbone_config = Dinov2Config.from_pretrained(
+            "facebook/dinov2-large", out_indices=[21, 22, 23, 24], apply_layernorm=True, reshape_hidden_states=False
+        )
+        fusion_hidden_size = 256
+        neck_hidden_sizes = [256, 512, 1024, 1024]
+    else:
+        raise NotImplementedError("To do")
+
+    config = DepthAnythingConfig(
+        reassemble_hidden_size=backbone_config.hidden_size,
+        patch_size=backbone_config.patch_size,
+        backbone_config=backbone_config,
+        fusion_hidden_size=fusion_hidden_size,
+        neck_hidden_sizes=neck_hidden_sizes,
+    )
+
+    return config
+
+
+def create_rename_keys(config):
+    rename_keys = []
+
+    # fmt: off
+    # stem
+    rename_keys.append(("pretrained.cls_token", "backbone.embeddings.cls_token"))
+    rename_keys.append(("pretrained.mask_token", "backbone.embeddings.mask_token"))
+    rename_keys.append(("pretrained.pos_embed", "backbone.embeddings.position_embeddings"))
+    rename_keys.append(("pretrained.patch_embed.proj.weight", "backbone.embeddings.patch_embeddings.projection.weight"))
+    rename_keys.append(("pretrained.patch_embed.proj.bias", "backbone.embeddings.patch_embeddings.projection.bias"))
+
+    # Transfomer encoder
+    for i in range(config.backbone_config.num_hidden_layers):
+        rename_keys.append((f"pretrained.blocks.{i}.ls1.gamma", f"backbone.encoder.layer.{i}.layer_scale1.lambda1"))
+        rename_keys.append((f"pretrained.blocks.{i}.ls2.gamma", f"backbone.encoder.layer.{i}.layer_scale2.lambda1"))
+        rename_keys.append((f"pretrained.blocks.{i}.norm1.weight", f"backbone.encoder.layer.{i}.norm1.weight"))
+        rename_keys.append((f"pretrained.blocks.{i}.norm1.bias", f"backbone.encoder.layer.{i}.norm1.bias"))
+        rename_keys.append((f"pretrained.blocks.{i}.norm2.weight", f"backbone.encoder.layer.{i}.norm2.weight"))
+        rename_keys.append((f"pretrained.blocks.{i}.norm2.bias", f"backbone.encoder.layer.{i}.norm2.bias"))
+        rename_keys.append((f"pretrained.blocks.{i}.mlp.fc1.weight", f"backbone.encoder.layer.{i}.mlp.fc1.weight"))
+        rename_keys.append((f"pretrained.blocks.{i}.mlp.fc1.bias", f"backbone.encoder.layer.{i}.mlp.fc1.bias"))
+        rename_keys.append((f"pretrained.blocks.{i}.mlp.fc2.weight", f"backbone.encoder.layer.{i}.mlp.fc2.weight"))
+        rename_keys.append((f"pretrained.blocks.{i}.mlp.fc2.bias", f"backbone.encoder.layer.{i}.mlp.fc2.bias"))
+        rename_keys.append((f"pretrained.blocks.{i}.attn.proj.weight", f"backbone.encoder.layer.{i}.attention.output.dense.weight"))
+        rename_keys.append((f"pretrained.blocks.{i}.attn.proj.bias", f"backbone.encoder.layer.{i}.attention.output.dense.bias"))
+
+    # Head
+    rename_keys.append(("pretrained.norm.weight", "backbone.layernorm.weight"))
+    rename_keys.append(("pretrained.norm.bias", "backbone.layernorm.bias"))
+
+    # activation postprocessing (readout projections + resize blocks)
+    # Depth Anything does not use CLS token => readout_projects not required
+
+    for i in range(4):
+        rename_keys.append((f"depth_head.projects.{i}.weight", f"neck.reassemble_stage.layers.{i}.projection.weight"))
+        rename_keys.append((f"depth_head.projects.{i}.bias", f"neck.reassemble_stage.layers.{i}.projection.bias"))
+
+        if i != 2:
+            rename_keys.append((f"depth_head.resize_layers.{i}.weight", f"neck.reassemble_stage.layers.{i}.resize.weight"))
+            rename_keys.append((f"depth_head.resize_layers.{i}.bias", f"neck.reassemble_stage.layers.{i}.resize.bias"))
+
+    # refinenet (tricky here)
+    mapping = {1:3, 2:2, 3:1, 4:0}
+
+    for i in range(1, 5):
+        j = mapping[i]
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.out_conv.weight", f"neck.fusion_stage.layers.{j}.projection.weight"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.out_conv.bias", f"neck.fusion_stage.layers.{j}.projection.bias"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit1.conv1.weight", f"neck.fusion_stage.layers.{j}.residual_layer1.convolution1.weight"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit1.conv1.bias", f"neck.fusion_stage.layers.{j}.residual_layer1.convolution1.bias"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit1.conv2.weight", f"neck.fusion_stage.layers.{j}.residual_layer1.convolution2.weight"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit1.conv2.bias", f"neck.fusion_stage.layers.{j}.residual_layer1.convolution2.bias"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit2.conv1.weight", f"neck.fusion_stage.layers.{j}.residual_layer2.convolution1.weight"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit2.conv1.bias", f"neck.fusion_stage.layers.{j}.residual_layer2.convolution1.bias"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit2.conv2.weight", f"neck.fusion_stage.layers.{j}.residual_layer2.convolution2.weight"))
+        rename_keys.append((f"depth_head.scratch.refinenet{i}.resConfUnit2.conv2.bias", f"neck.fusion_stage.layers.{j}.residual_layer2.convolution2.bias"))
+
+    # scratch convolutions
+    for i in range(4):
+        rename_keys.append((f"depth_head.scratch.layer{i+1}_rn.weight", f"neck.convs.{i}.weight"))
+
+    # head
+    rename_keys.append(("depth_head.scratch.output_conv1.weight", "head.conv1.weight"))
+    rename_keys.append(("depth_head.scratch.output_conv1.bias", "head.conv1.bias"))
+    rename_keys.append(("depth_head.scratch.output_conv2.0.weight", "head.conv2.weight"))
+    rename_keys.append(("depth_head.scratch.output_conv2.0.bias", "head.conv2.bias"))
+    rename_keys.append(("depth_head.scratch.output_conv2.2.weight", "head.conv3.weight"))
+    rename_keys.append(("depth_head.scratch.output_conv2.2.bias", "head.conv3.bias"))
+
+    return rename_keys
+
+
+# we split up the matrix of each encoder layer into queries, keys and values
+def read_in_q_k_v(state_dict, config):
+    hidden_size = config.backbone_config.hidden_size
+    for i in range(config.backbone_config.num_hidden_layers):
+        # read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
+        in_proj_weight = state_dict.pop(f"pretrained.blocks.{i}.attn.qkv.weight")
+        in_proj_bias = state_dict.pop(f"pretrained.blocks.{i}.attn.qkv.bias")
+        # next, add query, keys and values (in that order) to the state dict
+        state_dict[f"backbone.encoder.layer.{i}.attention.attention.query.weight"] = in_proj_weight[:hidden_size, :]
+        state_dict[f"backbone.encoder.layer.{i}.attention.attention.query.bias"] = in_proj_bias[:hidden_size]
+        state_dict[f"backbone.encoder.layer.{i}.attention.attention.key.weight"] = in_proj_weight[
+            hidden_size : hidden_size * 2, :
+        ]
+        state_dict[f"backbone.encoder.layer.{i}.attention.attention.key.bias"] = in_proj_bias[
+            hidden_size : hidden_size * 2
+        ]
+        state_dict[f"backbone.encoder.layer.{i}.attention.attention.value.weight"] = in_proj_weight[-hidden_size:, :]
+        state_dict[f"backbone.encoder.layer.{i}.attention.attention.value.bias"] = in_proj_bias[-hidden_size:]
+
+
+def rename_key(dct, old, new):
+    val = dct.pop(old)
+    dct[new] = val
+
+
+# We will verify our results on an image of cute cats
+def prepare_img():
+    url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+    im = Image.open(requests.get(url, stream=True).raw)
+    return im
+
+
+name_to_checkpoint = {
+    "depth-anything-small": "depth_anything_vits14.pth",
+    "depth-anything-base": "depth_anything_vitb14.pth",
+    "depth-anything-large": "depth_anything_vitl14.pth",
+}
+
+
+@torch.no_grad()
+def convert_dpt_checkpoint(model_name, pytorch_dump_folder_path, push_to_hub, verify_logits):
+    """
+    Copy/paste/tweak model's weights to our DPT structure.
+    """
+
+    # define DPT configuration
+    config = get_dpt_config(model_name)
+
+    model_name_to_filename = {
+        "depth-anything-small": "depth_anything_vits14.pth",
+        "depth-anything-base": "depth_anything_vitb14.pth",
+        "depth-anything-large": "depth_anything_vitl14.pth",
+    }
+
+    # load original state_dict
+    filename = model_name_to_filename[model_name]
+    filepath = hf_hub_download(
+        repo_id="LiheYoung/Depth-Anything", filename=f"checkpoints/{filename}", repo_type="space"
+    )
+    state_dict = torch.load(filepath, map_location="cpu")
+    # rename keys
+    rename_keys = create_rename_keys(config)
+    for src, dest in rename_keys:
+        rename_key(state_dict, src, dest)
+    # read in qkv matrices
+    read_in_q_k_v(state_dict, config)
+
+    # load HuggingFace model
+    model = DepthAnythingForDepthEstimation(config)
+    model.load_state_dict(state_dict)
+    model.eval()
+
+    processor = DPTImageProcessor(
+        do_resize=True,
+        size={"height": 518, "width": 518},
+        ensure_multiple_of=14,
+        keep_aspect_ratio=True,
+        do_rescale=True,
+        do_normalize=True,
+        image_mean=[0.485, 0.456, 0.406],
+        image_std=[0.229, 0.224, 0.225],
+    )
+
+    url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+    image = Image.open(requests.get(url, stream=True).raw)
+
+    pixel_values = processor(image, return_tensors="pt").pixel_values
+
+    # Verify forward pass
+    with torch.no_grad():
+        outputs = model(pixel_values)
+        predicted_depth = outputs.predicted_depth
+
+    print("Shape of predicted depth:", predicted_depth.shape)
+    print("First values:", predicted_depth[0, :3, :3])
+
+    # assert logits
+    if verify_logits:
+        expected_shape = torch.Size([1, 518, 686])
+        if model_name == "depth-anything-small":
+            expected_slice = torch.tensor(
+                [[8.8204, 8.6468, 8.6195], [8.3313, 8.6027, 8.7526], [8.6526, 8.6866, 8.7453]],
+            )
+        elif model_name == "depth-anything-base":
+            expected_slice = torch.tensor(
+                [[26.3997, 26.3004, 26.3928], [26.2260, 26.2092, 26.3427], [26.0719, 26.0483, 26.1254]],
+            )
+        elif model_name == "depth-anything-large":
+            expected_slice = torch.tensor(
+                [[87.9968, 87.7493, 88.2704], [87.1927, 87.6611, 87.3640], [86.7789, 86.9469, 86.7991]]
+            )
+        else:
+            raise ValueError("Not supported")
+
+        assert predicted_depth.shape == torch.Size(expected_shape)
+        assert torch.allclose(predicted_depth[0, :3, :3], expected_slice, atol=1e-6)
+        print("Looks ok!")
+
+    if pytorch_dump_folder_path is not None:
+        Path(pytorch_dump_folder_path).mkdir(exist_ok=True)
+        print(f"Saving model and processor to {pytorch_dump_folder_path}")
+        model.save_pretrained(pytorch_dump_folder_path)
+        processor.save_pretrained(pytorch_dump_folder_path)
+
+    if push_to_hub:
+        print("Pushing model and processor to hub...")
+        model.push_to_hub(repo_id=f"LiheYoung/{model_name}-hf")
+        processor.push_to_hub(repo_id=f"LiheYoung/{model_name}-hf")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--model_name",
+        default="depth-anything-small",
+        type=str,
+        choices=name_to_checkpoint.keys(),
+        help="Name of the model you'd like to convert.",
+    )
+    parser.add_argument(
+        "--pytorch_dump_folder_path",
+        default=None,
+        type=str,
+        help="Path to the output PyTorch model directory.",
+    )
+    parser.add_argument(
+        "--push_to_hub",
+        action="store_true",
+        help="Whether to push the model to the hub after conversion.",
+    )
+    parser.add_argument(
+        "--verify_logits",
+        action="store_false",
+        required=False,
+        help="Whether to verify the logits after conversion.",
+    )
+
+    args = parser.parse_args()
+    convert_dpt_checkpoint(args.model_name, args.pytorch_dump_folder_path, args.push_to_hub, args.verify_logits)

src/transformers/models/depth_anything/modeling_depth_anything.py

+# coding=utf-8
+# Copyright 2024 TikTok and The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch Depth Anything model."""
+
+
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+
+from ...file_utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+from ...modeling_outputs import DepthEstimatorOutput
+from ...modeling_utils import PreTrainedModel
+from ...utils import logging
+from ..auto import AutoBackbone
+from .configuration_depth_anything import DepthAnythingConfig
+
+
+logger = logging.get_logger(__name__)
+
+# General docstring
+_CONFIG_FOR_DOC = "DepthAnythingConfig"
+
+DEPTH_ANYTHING_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "LiheYoung/depth-anything-small-hf",
+    # See all Depth Anything models at https://huggingface.co/models?filter=depth_anything
+]
+
+
+DEPTH_ANYTHING_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it
+    as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config ([`DepthAnythingConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+DEPTH_ANYTHING_INPUTS_DOCSTRING = r"""
+    Args:
+        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
+            Pixel values. Pixel values can be obtained using [`AutoImageProcessor`]. See [`DPTImageProcessor.__call__`]
+            for details.
+
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+class DepthAnythingReassembleLayer(nn.Module):
+    def __init__(self, config, channels, factor):
+        super().__init__()
+        self.projection = nn.Conv2d(in_channels=config.reassemble_hidden_size, out_channels=channels, kernel_size=1)
+
+        # up/down sampling depending on factor
+        if factor > 1:
+            self.resize = nn.ConvTranspose2d(channels, channels, kernel_size=factor, stride=factor, padding=0)
+        elif factor == 1:
+            self.resize = nn.Identity()
+        elif factor < 1:
+            # so should downsample
+            self.resize = nn.Conv2d(channels, channels, kernel_size=3, stride=int(1 / factor), padding=1)
+
+    # Copied from transformers.models.dpt.modeling_dpt.DPTReassembleLayer.forward
+    def forward(self, hidden_state):
+        hidden_state = self.projection(hidden_state)
+        hidden_state = self.resize(hidden_state)
+
+        return hidden_state
+
+
+class DepthAnythingReassembleStage(nn.Module):
+    """
+    This class reassembles the hidden states of the backbone into image-like feature representations at various
+    resolutions.
+
+    This happens in 3 stages:
+    1. Take the patch embeddings and reshape them to image-like feature representations.
+    2. Project the channel dimension of the hidden states according to `config.neck_hidden_sizes`.
+    3. Resizing the spatial dimensions (height, width).
+
+    Args:
+        config (`[DepthAnythingConfig]`):
+            Model configuration class defining the model architecture.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.config = config
+        self.layers = nn.ModuleList()
+        for channels, factor in zip(config.neck_hidden_sizes, config.reassemble_factors):
+            self.layers.append(DepthAnythingReassembleLayer(config, channels=channels, factor=factor))
+
+    def forward(self, hidden_states: List[torch.Tensor], patch_height=None, patch_width=None) -> List[torch.Tensor]:
+        """
+        Args:
+            hidden_states (`List[torch.FloatTensor]`, each of shape `(batch_size, sequence_length + 1, hidden_size)`):
+                List of hidden states from the backbone.
+        """
+        out = []
+
+        for i, hidden_state in enumerate(hidden_states):
+            # reshape to (batch_size, num_channels, height, width)
+            hidden_state = hidden_state[:, 1:]
+            batch_size, _, num_channels = hidden_state.shape
+            hidden_state = hidden_state.reshape(batch_size, patch_height, patch_width, num_channels)
+            hidden_state = hidden_state.permute(0, 3, 1, 2).contiguous()
+            hidden_state = self.layers[i](hidden_state)
+            out.append(hidden_state)
+
+        return out
+
+
+class DepthAnythingPreActResidualLayer(nn.Module):
+    """
+    ResidualConvUnit, pre-activate residual unit.
+
+    Args:
+        config (`[DepthAnythingConfig]`):
+            Model configuration class defining the model architecture.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.activation1 = nn.ReLU()
+        self.convolution1 = nn.Conv2d(
+            config.fusion_hidden_size,
+            config.fusion_hidden_size,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias=True,
+        )
+
+        self.activation2 = nn.ReLU()
+        self.convolution2 = nn.Conv2d(
+            config.fusion_hidden_size,
+            config.fusion_hidden_size,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias=True,
+        )
+
+    def forward(self, hidden_state: torch.Tensor) -> torch.Tensor:
+        residual = hidden_state
+        hidden_state = self.activation1(hidden_state)
+        hidden_state = self.convolution1(hidden_state)
+        hidden_state = self.activation2(hidden_state)
+        hidden_state = self.convolution2(hidden_state)
+
+        return hidden_state + residual
+
+
+class DepthAnythingFeatureFusionLayer(nn.Module):
+    """Feature fusion layer, merges feature maps from different stages.
+
+    Args:
+        config (`[DepthAnythingConfig]`):
+            Model configuration class defining the model architecture.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.projection = nn.Conv2d(config.fusion_hidden_size, config.fusion_hidden_size, kernel_size=1, bias=True)
+
+        self.residual_layer1 = DepthAnythingPreActResidualLayer(config)
+        self.residual_layer2 = DepthAnythingPreActResidualLayer(config)
+
+    def forward(self, hidden_state, residual=None, size=None):
+        if residual is not None:
+            if hidden_state.shape != residual.shape:
+                residual = nn.functional.interpolate(
+                    residual, size=(hidden_state.shape[2], hidden_state.shape[3]), mode="bilinear", align_corners=False
+                )
+            hidden_state = hidden_state + self.residual_layer1(residual)
+
+        hidden_state = self.residual_layer2(hidden_state)
+
+        modifier = {"scale_factor": 2} if size is None else {"size": size}
+
+        hidden_state = nn.functional.interpolate(
+            hidden_state,
+            **modifier,
+            mode="bilinear",
+            align_corners=True,
+        )
+        hidden_state = self.projection(hidden_state)
+
+        return hidden_state
+
+
+class DepthAnythingFeatureFusionStage(nn.Module):
+    # Copied from transformers.models.dpt.modeling_dpt.DPTFeatureFusionStage.__init__ with DPT->DepthAnything
+    def __init__(self, config):
+        super().__init__()
+        self.layers = nn.ModuleList()
+        for _ in range(len(config.neck_hidden_sizes)):
+            self.layers.append(DepthAnythingFeatureFusionLayer(config))
+
+    def forward(self, hidden_states, size=None):
+        # reversing the hidden_states, we start from the last
+        hidden_states = hidden_states[::-1]
+
+        fused_hidden_states = []
+        # first layer only uses the last hidden_state
+        size = hidden_states[1].shape[2:]
+        fused_hidden_state = self.layers[0](hidden_states[0], size=size)
+        fused_hidden_states.append(fused_hidden_state)
+
+        # looping from the last layer to the second
+        for idx, (hidden_state, layer) in enumerate(zip(hidden_states[1:], self.layers[1:])):
+            size = hidden_states[1:][idx + 1].shape[2:] if idx != (len(hidden_states[1:]) - 1) else None
+
+            fused_hidden_state = layer(fused_hidden_state, hidden_state, size=size)
+
+            fused_hidden_states.append(fused_hidden_state)
+
+        return fused_hidden_states
+
+
+# Copied from transformers.models.dpt.modeling_dpt.DPTPreTrainedModel with DPT->DepthAnything,dpt->depth_anything
+class DepthAnythingPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = DepthAnythingConfig
+    base_model_prefix = "depth_anything"
+    main_input_name = "pixel_values"
+    supports_gradient_checkpointing = True
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv2d, nn.ConvTranspose2d)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+
+class DepthAnythingNeck(nn.Module):
+    """
+    DepthAnythingNeck. A neck is a module that is normally used between the backbone and the head. It takes a list of tensors as
+    input and produces another list of tensors as output. For DepthAnything, it includes 2 stages:
+
+    * DepthAnythingReassembleStage
+    * DepthAnythingFeatureFusionStage.
+
+    Args:
+        config (dict): config dict.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+
+        self.reassemble_stage = DepthAnythingReassembleStage(config)
+
+        self.convs = nn.ModuleList()
+        for channel in config.neck_hidden_sizes:
+            self.convs.append(nn.Conv2d(channel, config.fusion_hidden_size, kernel_size=3, padding=1, bias=False))
+
+        # fusion
+        self.fusion_stage = DepthAnythingFeatureFusionStage(config)
+
+    def forward(self, hidden_states: List[torch.Tensor], patch_height=None, patch_width=None) -> List[torch.Tensor]:
+        """
+        Args:
+            hidden_states (`List[torch.FloatTensor]`, each of shape `(batch_size, sequence_length, hidden_size)` or `(batch_size, hidden_size, height, width)`):
+                List of hidden states from the backbone.
+        """
+        if not isinstance(hidden_states, (tuple, list)):
+            raise ValueError("hidden_states should be a tuple or list of tensors")
+
+        if len(hidden_states) != len(self.config.neck_hidden_sizes):
+            raise ValueError("The number of hidden states should be equal to the number of neck hidden sizes.")
+
+        # postprocess hidden states
+        hidden_states = self.reassemble_stage(hidden_states, patch_height, patch_width)
+
+        features = [self.convs[i](feature) for i, feature in enumerate(hidden_states)]
+
+        # fusion blocks
+        output = self.fusion_stage(features)
+
+        return output
+
+
+class DepthAnythingDepthEstimationHead(nn.Module):
+    """
+    Output head consisting of 3 convolutional layers. It progressively halves the feature dimension and upsamples
+    the predictions to the input resolution after the first convolutional layer (details can be found in the DPT paper's
+    supplementary material).
+    """
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.head_in_index = config.head_in_index
+        self.patch_size = config.patch_size
+
+        features = config.fusion_hidden_size
+        self.conv1 = nn.Conv2d(features, features // 2, kernel_size=3, stride=1, padding=1)
+        self.conv2 = nn.Conv2d(features // 2, config.head_hidden_size, kernel_size=3, stride=1, padding=1)
+        self.activation1 = nn.ReLU()
+        self.conv3 = nn.Conv2d(config.head_hidden_size, 1, kernel_size=1, stride=1, padding=0)
+        self.activation2 = nn.ReLU()
+
+    def forward(self, hidden_states: List[torch.Tensor], patch_height, patch_width) -> torch.Tensor:
+        hidden_states = hidden_states[self.head_in_index]
+
+        predicted_depth = self.conv1(hidden_states)
+        predicted_depth = nn.functional.interpolate(
+            predicted_depth,
+            (int(patch_height * self.patch_size), int(patch_width * self.patch_size)),
+            mode="bilinear",
+            align_corners=True,
+        )
+        predicted_depth = self.conv2(predicted_depth)
+        predicted_depth = self.activation1(predicted_depth)
+        predicted_depth = self.conv3(predicted_depth)
+        predicted_depth = self.activation2(predicted_depth)
+        predicted_depth = predicted_depth.squeeze(dim=1)  # shape (batch_size, height, width)
+
+        return predicted_depth
+
+
+@add_start_docstrings(
+    """
+    Depth Anything Model with a depth estimation head on top (consisting of 3 convolutional layers) e.g. for KITTI, NYUv2.
+    """,
+    DEPTH_ANYTHING_START_DOCSTRING,
+)
+class DepthAnythingForDepthEstimation(DepthAnythingPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.backbone = AutoBackbone.from_config(config.backbone_config)
+        self.neck = DepthAnythingNeck(config)
+        self.head = DepthAnythingDepthEstimationHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(DEPTH_ANYTHING_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=DepthEstimatorOutput, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        pixel_values: torch.FloatTensor,
+        labels: Optional[torch.LongTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], DepthEstimatorOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, height, width)`, *optional*):
+            Ground truth depth estimation maps for computing the loss.
+
+        Returns:
+
+        Examples:
+        ```python
+        >>> from transformers import AutoImageProcessor, AutoModelForDepthEstimation
+        >>> import torch
+        >>> import numpy as np
+        >>> from PIL import Image
+        >>> import requests
+
+        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+        >>> image = Image.open(requests.get(url, stream=True).raw)
+
+        >>> image_processor = AutoImageProcessor.from_pretrained("LiheYoung/depth-anything-small-hf")
+        >>> model = AutoModelForDepthEstimation.from_pretrained("LiheYoung/depth-anything-small-hf")
+
+        >>> # prepare image for the model
+        >>> inputs = image_processor(images=image, return_tensors="pt")
+
+        >>> with torch.no_grad():
+        ...     outputs = model(**inputs)
+        ...     predicted_depth = outputs.predicted_depth
+
+        >>> # interpolate to original size
+        >>> prediction = torch.nn.functional.interpolate(
+        ...     predicted_depth.unsqueeze(1),
+        ...     size=image.size[::-1],
+        ...     mode="bicubic",
+        ...     align_corners=False,
+        ... )
+
+        >>> # visualize the prediction
+        >>> output = prediction.squeeze().cpu().numpy()
+        >>> formatted = (output * 255 / np.max(output)).astype("uint8")
+        >>> depth = Image.fromarray(formatted)
+        ```"""
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+
+        outputs = self.backbone.forward_with_filtered_kwargs(
+            pixel_values, output_hidden_states=output_hidden_states, output_attentions=output_attentions
+        )
+        hidden_states = outputs.feature_maps
+
+        _, _, height, width = pixel_values.shape
+        patch_size = self.config.patch_size
+        patch_height = height // patch_size
+        patch_width = width // patch_size
+
+        hidden_states = self.neck(hidden_states, patch_height, patch_width)
+
+        predicted_depth = self.head(hidden_states, patch_height, patch_width)
+
+        loss = None
+        if labels is not None:
+            raise NotImplementedError("Training is not implemented yet")
+
+        if not return_dict:
+            if output_hidden_states:
+                output = (predicted_depth,) + outputs[1:]
+            else:
+                output = (predicted_depth,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return DepthEstimatorOutput(
+            loss=loss,
+            predicted_depth=predicted_depth,
+            hidden_states=outputs.hidden_states if output_hidden_states else None,
+            attentions=outputs.attentions,
+        )