Adding SegGPT (#27735)

* First commit

* Improvements

* More improvements

* Converted original checkpoint to HF checkpoint

* Fix style

* Fixed forward

* More improvements

* More improvements

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Remove asserts

* Remove unnecessary attributes

* Changed model name to camel case

* Improve forward doc

* Improve tests

* More improvements

* Fix copies

* Fix doc

* Make SegGptImageProcessor more flexible

* Added few-shot test

* Fix style

* Update READMEs and docs

* Update READMEs

* Make inputs required

* Add SegGptForImageSegmentation

* Make tests pass

* Rename to out_indicies

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Fixed naming convention

* Copying SegGptMlp from modeling_sam.py

* Some minor improvements

* Remove mlp_ratio

* Fix docstrings

* Fixed docstring match

* Objects defined before use

* Storing only patch_size and beta for SegGptLoss

* removed _prepare_inputs method

* Removed modified from headers

* Renamed to output_indicies

* Removed unnecessary einsums

* Update tests/models/seggpt/test_modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update tests/models/seggpt/test_modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update tests/models/seggpt/test_modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Fixing issues

* Raise error as soon as possible

* More fixes

* Fix merge

* Added palette to SegGptImageProcessor

* Fixed typo

* Fixed shape typo

* Added permute before doing palette to class mapping

* Fixed style

* Fixed and added tests

* Fixed docstrings

* Matching SegFormer API for post_processing_semantic_segmentation

* Fixed copies

* Fixed SegGptImageProcessor to handle both binary and RGB masks

* Updated docstrings of SegGptImageProcessor

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update docs/source/en/model_doc/seggpt.md
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/configuration_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/convert_seggpt_to_hf.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update tests/models/seggpt/test_image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update tests/models/seggpt/test_modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Object definitions above & fix style

* Renamed output_indices to intermediate_feature_indices

* Removed unnecessary check on bool_masked_pos

* Loss first in the outputs

* Added validation for do_normalize

* Improved SegGptImageProcessor and added new tests

* Added comment

* Added docstrings to SegGptLoss

* Reimplemented ensemble condition logic in SegGptEncoder

* Update src/transformers/models/seggpt/__init__.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update src/transformers/models/seggpt/convert_seggpt_to_hf.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Update src/transformers/models/seggpt/configuration_seggpt.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Updated docstrings to use post_process_semantic_segmentation

* Fixed typo on docstrings

* moved pixel values test to test_image_processing_seggpt

* Addressed comments

* Update src/transformers/models/seggpt/configuration_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/image_processing_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/configuration_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Updated docstrings for SegGptLoss

* Address comments

* Added SegGpt example to model docs

* Update src/transformers/models/seggpt/modeling_seggpt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* moved patchify and unpatchify

* Rename checkpoint

* Renamed intermediate_features to intermediate_hidden_states for consistency

* Update src/transformers/models/seggpt/configuration_seggpt.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* Replaced post_process_masks for post_process_semantic_segmentation in the docs

---------
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
Co-authored-by: Niels <niels.rogge1@gmail.com>
Co-authored-by: Eduardo Pacheco <eduardo.pacheco@limehome.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

README.md

@@ -482,6 +482,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (from Beijing Academy of Artificial Intelligence (BAAI)) released with the paper [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) by Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

README_es.md

@@ -455,6 +455,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (from Beijing Academy of Artificial Intelligence (BAAI) released with the paper [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) by Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

README_fr.md

@@ -476,6 +476,7 @@ Nombre actuel de points de contrôle : ![](https://img.shields.io/endpoint?url=h
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (de Meta AI) a été publié dans l'article [SeamlessM4T — Traduction multimodale et massivement multilingue](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) par l'équipe de communication transparente.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (de Meta AI) a été publié dans l'article [Seamless: Traduction de la parole multilingue, expressive et en continu](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) par l'équipe de communication transparente.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (de NVIDIA) a été publié dans l'article [SegFormer : Conception simple et efficace pour la segmentation sémantique avec des transformateurs](https://arxiv.org/abs/2105.15203) par Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (de Beijing Academy of Artificial Intelligence (BAAI) publié dans l'article [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) parXinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (de Meta AI) a été publié dans l'article [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) par Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (de ASAPP) a été publié dans l'article [Compromis entre performances et efficacité dans l'entraînement non supervisé pour la reconnaissance vocale](https://arxiv.org/abs/2109.06870) par Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (de ASAPP) a été publié dans l'article [Compromis entre performances et efficacité dans l'entraînement non supervisé pour la reconnaissance vocale](https://arxiv.org/abs/2109.06870) par Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

README_hd.md

@@ -429,6 +429,7 @@ conda install conda-forge::transformers
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (Beijing Academy of Artificial Intelligence (BAAI से) Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang. द्वाराअनुसंधान पत्र [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) के साथ जारी किया गया
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (Meta AI से) Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick. द्वाराअनुसंधान पत्र [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) के साथ जारी किया गया
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP से) साथ देने वाला पेपर [भाषण पहचान के लिए अनसुपरवाइज्ड प्री-ट्रेनिंग में परफॉर्मेंस-एफिशिएंसी ट्रेड-ऑफ्स](https://arxiv.org/abs/2109.06870) फेलिक्स वू, क्वांगयुन किम, जिंग पैन, क्यू हान, किलियन क्यू. वेनबर्गर, योव आर्टज़ी द्वारा।
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (ASAPP से) साथ में पेपर [भाषण पहचान के लिए अनसुपरवाइज्ड प्री-ट्रेनिंग में परफॉर्मेंस-एफिशिएंसी ट्रेड-ऑफ्स](https://arxiv.org/abs/2109.06870) फेलिक्स वू, क्वांगयुन किम, जिंग पैन, क्यू हान, किलियन क्यू. वेनबर्गर, योआव आर्टज़ी द्वारा पोस्ट किया गया।

README_ja.md

@@ -489,6 +489,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (NVIDIA から) Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo から公開された研究論文: [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203)
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (Beijing Academy of Artificial Intelligence (BAAI から) Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang. から公開された研究論文 [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284)
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (Meta AI から) Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick. から公開された研究論文 [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf)
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP から) Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi から公開された研究論文: [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870)
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (ASAPP から) Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi から公開された研究論文: [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870)

README_ko.md

@@ -404,6 +404,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (NVIDIA 에서) Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 의 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 논문과 함께 발표했습니다.
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (Beijing Academy of Artificial Intelligence (BAAI 에서 제공)은 Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang.의 [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284)논문과 함께 발표했습니다.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (Meta AI 에서 제공)은 Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.의 [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf)논문과 함께 발표했습니다.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP 에서) Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 의 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 논문과 함께 발표했습니다.
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (ASAPP 에서) Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 의 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 논문과 함께 발표했습니다.

README_zh-hans.md

@@ -428,6 +428,7 @@ conda install conda-forge::transformers
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (来自 NVIDIA) 伴随论文 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 由 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 发布。
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (来自 Beijing Academy of Artificial Intelligence (BAAI) 伴随论文 [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) 由 Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang 发布。
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (来自 Meta AI) 伴随论文 [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) 由 Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick 发布。
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。

README_zh-hant.md

@@ -440,6 +440,7 @@ conda install conda-forge::transformers
 1. **[SeamlessM4T](https://huggingface.co/docs/transformers/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SeamlessM4Tv2](https://huggingface.co/docs/transformers/model_doc/seamless_m4t_v2)** (from Meta AI) released with the paper [Seamless: Multilingual Expressive and Streaming Speech Translation](https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SegGPT](https://huggingface.co/docs/transformers/main/model_doc/seggpt)** (from Beijing Academy of Artificial Intelligence (BAAI) released with the paper [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) by Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
 1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

docs/source/en/_toctree.yml

@@ -583,6 +583,8 @@
         title: ResNet
       - local: model_doc/segformer
         title: SegFormer
+      - local: model_doc/seggpt
+        title: SegGpt
       - local: model_doc/swiftformer
         title: SwiftFormer
       - local: model_doc/swin

docs/source/en/index.md

@@ -251,6 +251,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                  [SeamlessM4T](model_doc/seamless_m4t)                   |       ✅        |         ❌         |      ❌      |
 |                [SeamlessM4Tv2](model_doc/seamless_m4t_v2)                |       ✅        |         ❌         |      ❌      |
 |                     [SegFormer](model_doc/segformer)                     |       ✅        |         ✅         |      ❌      |
+|                        [SegGPT](model_doc/seggpt)                        |       ✅        |         ❌         |      ❌      |
 |                           [SEW](model_doc/sew)                           |       ✅        |         ❌         |      ❌      |
 |                         [SEW-D](model_doc/sew-d)                         |       ✅        |         ❌         |      ❌      |
 |                        [SigLIP](model_doc/siglip)                        |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/seggpt.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.
+
+-->
+
+# SegGPT
+
+## Overview
+
+The SegGPT model was proposed in [SegGPT: Segmenting Everything In Context](https://arxiv.org/abs/2304.03284) by Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, Tiejun Huang. SegGPT employs a decoder-only Transformer that can generate a segmentation mask given an input image, a prompt image and its corresponding prompt mask. The model achieves remarkable one-shot results with 56.1 mIoU on COCO-20 and 85.6 mIoU on FSS-1000.
+
+The abstract from the paper is the following:
+
+*We present SegGPT, a generalist model for segmenting everything in context. We unify various segmentation tasks into a generalist in-context learning framework that accommodates different kinds of segmentation data by transforming them into the same format of images. The training of SegGPT is formulated as an in-context coloring problem with random color mapping for each data sample. The objective is to accomplish diverse tasks according to the context, rather than relying on specific colors. After training, SegGPT can perform arbitrary segmentation tasks in images or videos via in-context inference, such as object instance, stuff, part, contour, and text. SegGPT is evaluated on a broad range of tasks, including few-shot semantic segmentation, video object segmentation, semantic segmentation, and panoptic segmentation. Our results show strong capabilities in segmenting in-domain and out-of*
+
+Tips:
+- One can use [`SegGptImageProcessor`] to prepare image input, prompt and mask to the model.
+- It's highly advisable to pass `num_labels` (not considering background) during preprocessing and postprocessing with [`SegGptImageProcessor`] for your use case.
+- When doing infenrece with [`SegGptForImageSegmentation`] if your `batch_size` is greater than 1 you can use feature ensemble across your images by passing `feature_ensemble=True` in the forward method.
+
+Here's how to use the model for one-shot semantic segmentation:
+
+```python
+import torch
+from datasets import load_dataset
+from transformers import SegGptImageProcessor, SegGptForImageSegmentation
+
+model_id = "BAAI/seggpt-vit-large"
+image_processor = SegGptImageProcessor.from_pretrained(checkpoint)
+model = SegGptForImageSegmentation.from_pretrained(checkpoint)
+
+dataset_id = "EduardoPacheco/FoodSeg103"
+ds = load_dataset(dataset_id, split="train")
+# Number of labels in FoodSeg103 (not including background)
+num_labels = 103
+
+image_input = ds[4]["image"]
+ground_truth = ds[4]["label"]
+image_prompt = ds[29]["image"]
+mask_prompt = ds[29]["label"]
+
+inputs = image_processor(
+    images=image_input, 
+    prompt_images=image_prompt,
+    prompt_masks=mask_prompt, 
+    num_labels=num_labels,
+    return_tensors="pt"
+)
+
+with torch.no_grad():
+    outputs = model(**inputs)
+
+target_sizes = [image_input.size[::-1]]
+mask = image_processor.post_process_semantic_segmentation(outputs, target_sizes, num_labels=num_labels)[0]
+```
+
+This model was contributed by [EduardoPacheco](https://huggingface.co/EduardoPacheco).
+The original code can be found [here]([(https://github.com/baaivision/Painter/tree/main)).
+
+
+## SegGptConfig
+
+[[autodoc]] SegGptConfig
+
+## SegGptImageProcessor
+
+[[autodoc]] SegGptImageProcessor
+    - preprocess
+    - post_process_semantic_segmentation
+
+## SegGptModel
+
+[[autodoc]] SegGptModel
+    - forward
+
+## SegGptForImageSegmentation
+
+[[autodoc]] SegGptForImageSegmentation
+    - forward
\ No newline at end of file

src/transformers/__init__.py

@@ -767,6 +767,7 @@ _import_structure = {
         "SeamlessM4Tv2Config",
     ],
     "models.segformer": ["SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "SegformerConfig"],
+    "models.seggpt": ["SEGGPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "SegGptConfig"],
     "models.sew": ["SEW_PRETRAINED_CONFIG_ARCHIVE_MAP", "SEWConfig"],
     "models.sew_d": ["SEW_D_PRETRAINED_CONFIG_ARCHIVE_MAP", "SEWDConfig"],
     "models.siglip": [
@@ -1316,6 +1317,7 @@ else:
     _import_structure["models.pvt"].extend(["PvtImageProcessor"])
     _import_structure["models.sam"].extend(["SamImageProcessor"])
     _import_structure["models.segformer"].extend(["SegformerFeatureExtractor", "SegformerImageProcessor"])
+    _import_structure["models.seggpt"].extend(["SegGptImageProcessor"])
     _import_structure["models.siglip"].append("SiglipImageProcessor")
     _import_structure["models.swin2sr"].append("Swin2SRImageProcessor")
     _import_structure["models.tvlt"].append("TvltImageProcessor")
@@ -3192,6 +3194,14 @@ else:
             "SegformerPreTrainedModel",
         ]
     )
+    _import_structure["models.seggpt"].extend(
+        [
+            "SEGGPT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "SegGptForImageSegmentation",
+            "SegGptModel",
+            "SegGptPreTrainedModel",
+        ]
+    )
     _import_structure["models.sew"].extend(
         [
             "SEW_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -5531,10 +5541,8 @@ if TYPE_CHECKING:
         SEAMLESS_M4T_V2_PRETRAINED_CONFIG_ARCHIVE_MAP,
         SeamlessM4Tv2Config,
     )
-    from .models.segformer import (
-        SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP,
-        SegformerConfig,
-    )
+    from .models.segformer import SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, SegformerConfig
+    from .models.seggpt import SEGGPT_PRETRAINED_CONFIG_ARCHIVE_MAP, SegGptConfig
     from .models.sew import SEW_PRETRAINED_CONFIG_ARCHIVE_MAP, SEWConfig
     from .models.sew_d import SEW_D_PRETRAINED_CONFIG_ARCHIVE_MAP, SEWDConfig
     from .models.siglip import (
@@ -6080,6 +6088,7 @@ if TYPE_CHECKING:
         from .models.pvt import PvtImageProcessor
         from .models.sam import SamImageProcessor
         from .models.segformer import SegformerFeatureExtractor, SegformerImageProcessor
+        from .models.seggpt import SegGptImageProcessor
         from .models.siglip import SiglipImageProcessor
         from .models.swin2sr import Swin2SRImageProcessor
         from .models.tvlt import TvltImageProcessor
@@ -7635,6 +7644,12 @@ if TYPE_CHECKING:
             SegformerModel,
             SegformerPreTrainedModel,
         )
+        from .models.seggpt import (
+            SEGGPT_PRETRAINED_MODEL_ARCHIVE_LIST,
+            SegGptForImageSegmentation,
+            SegGptModel,
+            SegGptPreTrainedModel,
+        )
         from .models.sew import (
             SEW_PRETRAINED_MODEL_ARCHIVE_LIST,
             SEWForCTC,

src/transformers/models/__init__.py

@@ -194,6 +194,7 @@ from . import (
     seamless_m4t,
     seamless_m4t_v2,
     segformer,
+    seggpt,
     sew,
     sew_d,
     siglip,

src/transformers/models/auto/configuration_auto.py

@@ -202,6 +202,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("seamless_m4t", "SeamlessM4TConfig"),
         ("seamless_m4t_v2", "SeamlessM4Tv2Config"),
         ("segformer", "SegformerConfig"),
+        ("seggpt", "SegGptConfig"),
         ("sew", "SEWConfig"),
         ("sew-d", "SEWDConfig"),
         ("siglip", "SiglipConfig"),
@@ -428,6 +429,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("seamless_m4t", "SEAMLESS_M4T_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("seamless_m4t_v2", "SEAMLESS_M4T_V2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("segformer", "SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("seggpt", "SEGGPT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("sew", "SEW_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("sew-d", "SEW_D_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("siglip", "SIGLIP_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -680,6 +682,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("seamless_m4t", "SeamlessM4T"),
         ("seamless_m4t_v2", "SeamlessM4Tv2"),
         ("segformer", "SegFormer"),
+        ("seggpt", "SegGPT"),
         ("sew", "SEW"),
         ("sew-d", "SEW-D"),
         ("siglip", "SigLIP"),

src/transformers/models/auto/image_processing_auto.py

@@ -98,6 +98,7 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("resnet", "ConvNextImageProcessor"),
         ("sam", "SamImageProcessor"),
         ("segformer", "SegformerImageProcessor"),
+        ("seggpt", "SegGptImageProcessor"),
         ("siglip", "SiglipImageProcessor"),
         ("swiftformer", "ViTImageProcessor"),
         ("swin", "ViTImageProcessor"),

src/transformers/models/auto/modeling_auto.py

@@ -193,6 +193,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("seamless_m4t", "SeamlessM4TModel"),
         ("seamless_m4t_v2", "SeamlessM4Tv2Model"),
         ("segformer", "SegformerModel"),
+        ("seggpt", "SegGptModel"),
         ("sew", "SEWModel"),
         ("sew-d", "SEWDModel"),
         ("siglip", "SiglipModel"),

src/transformers/models/seggpt/__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 ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available, is_vision_available
+
+
+_import_structure = {
+    "configuration_seggpt": ["SEGGPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "SegGptConfig", "SegGptOnnxConfig"]
+}
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_seggpt"] = [
+        "SEGGPT_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "SegGptModel",
+        "SegGptPreTrainedModel",
+        "SegGptForImageSegmentation",
+    ]
+
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["image_processing_seggpt"] = ["SegGptImageProcessor"]
+
+if TYPE_CHECKING:
+    from .configuration_seggpt import SEGGPT_PRETRAINED_CONFIG_ARCHIVE_MAP, SegGptConfig, SegGptOnnxConfig
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_seggpt import (
+            SEGGPT_PRETRAINED_MODEL_ARCHIVE_LIST,
+            SegGptForImageSegmentation,
+            SegGptModel,
+            SegGptPreTrainedModel,
+        )
+
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .image_processing_seggpt import SegGptImageProcessor
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)

src/transformers/models/seggpt/configuration_seggpt.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.
+""" SegGpt model configuration"""
+
+
+from ...configuration_utils import PretrainedConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+SEGGPT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "BAAI/seggpt-vit-large": "https://huggingface.co/BAAI/seggpt-vit-large/resolve/main/config.json",
+}
+
+
+class SegGptConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`SegGptModel`]. It is used to instantiate a SegGPT
+    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 SegGPT
+    [BAAI/seggpt-vit-large](https://huggingface.co/BAAI/seggpt-vit-large) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        hidden_size (`int`, *optional*, defaults to 1024):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 24):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"selu"` and `"gelu_new"` are supported.
+        hidden_dropout_prob (`float`, *optional*, defaults to 0.0):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the layer normalization layers.
+        image_size (`List[int]`, *optional*, defaults to `[896, 448]`):
+            The size (resolution) of each image.
+        patch_size (`int`, *optional*, defaults to 16):
+            The size (resolution) of each patch.
+        num_channels (`int`, *optional*, defaults to 3):
+            The number of input channels.
+        qkv_bias (`bool`, *optional*, defaults to `True`):
+            Whether to add a bias to the queries, keys and values.
+        mlp_dim (`int`, *optional*):
+            The dimensionality of the MLP layer in the Transformer encoder. If unset, defaults to
+            `hidden_size` * 4.
+        drop_path_rate (`float`, *optional*, defaults to 0.1):
+            The drop path rate for the dropout layers.
+        pretrain_image_size (`int`, *optional*, defaults to 224):
+            The pretrained size of the absolute position embeddings.
+        decoder_hidden_size (`int`, *optional*, defaults to 64):
+            Hidden size for decoder.
+        use_relative_position_embeddings (`bool`, *optional*, defaults to `True`):
+            Whether to use relative position embeddings in the attention layers.
+        merge_index (`int`, *optional*, defaults to 2):
+            The index of the encoder layer to merge the embeddings.
+        intermediate_hidden_state_indices (`List[int]`, *optional*, defaults to `[5, 11, 17, 23]`):
+            The indices of the encoder layers which we store as features for the decoder.
+        beta (`float`, *optional*, defaults to 0.01):
+            Regularization factor for SegGptLoss (smooth-l1 loss).
+
+    Example:
+
+    ```python
+    >>> from transformers import SegGptConfig, SegGptModel
+
+    >>> # Initializing a SegGPT seggpt-vit-large style configuration
+    >>> configuration = SegGptConfig()
+
+    >>> # Initializing a model (with random weights) from the seggpt-vit-large style configuration
+    >>> model = SegGptModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "seggpt"
+
+    def __init__(
+        self,
+        hidden_size=1024,
+        num_hidden_layers=24,
+        num_attention_heads=16,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.0,
+        initializer_range=0.02,
+        layer_norm_eps=1e-6,
+        image_size=[896, 448],
+        patch_size=16,
+        num_channels=3,
+        qkv_bias=True,
+        mlp_dim=None,
+        drop_path_rate=0.1,
+        pretrain_image_size=224,
+        decoder_hidden_size=64,
+        use_relative_position_embeddings=True,
+        merge_index=2,
+        intermediate_hidden_state_indices=[5, 11, 17, 23],
+        beta=0.01,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        if merge_index > min(intermediate_hidden_state_indices):
+            raise ValueError(
+                f"Merge index must be less than the minimum encoder output index, but got {merge_index=} and {intermediate_hidden_state_indices=}"
+            )
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.num_channels = num_channels
+        self.qkv_bias = qkv_bias
+        self.drop_path_rate = drop_path_rate
+        self.pretrain_image_size = pretrain_image_size
+        self.decoder_hidden_size = decoder_hidden_size
+        self.use_relative_position_embeddings = use_relative_position_embeddings
+        self.merge_index = merge_index
+        self.intermediate_hidden_state_indices = intermediate_hidden_state_indices
+        self.beta = beta
+        self.mlp_dim = int(hidden_size * 4) if mlp_dim is None else mlp_dim

src/transformers/models/seggpt/convert_seggpt_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 SegGPT checkpoints from the original repository.
+
+URL: https://github.com/baaivision/Painter/tree/main/SegGPT
+"""
+
+
+import argparse
+
+import requests
+import torch
+from PIL import Image
+
+from transformers import SegGptConfig, SegGptForImageSegmentation, SegGptImageProcessor
+from transformers.utils import logging
+
+
+logging.set_verbosity_info()
+logger = logging.get_logger(__name__)
+
+
+# here we list all keys to be renamed (original name on the left, our name on the right)
+def create_rename_keys(config):
+    rename_keys = []
+
+    # fmt: off
+
+    # rename embedding and its parameters
+    rename_keys.append(("patch_embed.proj.weight", "model.embeddings.patch_embeddings.projection.weight"))
+    rename_keys.append(("patch_embed.proj.bias", "model.embeddings.patch_embeddings.projection.bias"))
+    rename_keys.append(("mask_token", "model.embeddings.mask_token"))
+    rename_keys.append(("segment_token_x", "model.embeddings.segment_token_input"))
+    rename_keys.append(("segment_token_y", "model.embeddings.segment_token_prompt"))
+    rename_keys.append(("type_token_cls", "model.embeddings.type_token_semantic"))
+    rename_keys.append(("type_token_ins", "model.embeddings.type_token_instance"))
+    rename_keys.append(("pos_embed", "model.embeddings.position_embeddings"))
+
+    # rename decoder and other
+    rename_keys.append(("norm.weight", "model.encoder.layernorm.weight"))
+    rename_keys.append(("norm.bias", "model.encoder.layernorm.bias"))
+    rename_keys.append(("decoder_embed.weight", "decoder.decoder_embed.weight"))
+    rename_keys.append(("decoder_embed.bias", "decoder.decoder_embed.bias"))
+    rename_keys.append(("decoder_pred.0.weight", "decoder.decoder_pred.conv.weight"))
+    rename_keys.append(("decoder_pred.0.bias", "decoder.decoder_pred.conv.bias"))
+    rename_keys.append(("decoder_pred.1.weight", "decoder.decoder_pred.layernorm.weight"))
+    rename_keys.append(("decoder_pred.1.bias", "decoder.decoder_pred.layernorm.bias"))
+    rename_keys.append(("decoder_pred.3.weight", "decoder.decoder_pred.head.weight"))
+    rename_keys.append(("decoder_pred.3.bias", "decoder.decoder_pred.head.bias"))
+
+    # rename blocks
+    for i in range(config.num_hidden_layers):
+        rename_keys.append((f"blocks.{i}.attn.qkv.weight", f"model.encoder.layers.{i}.attention.qkv.weight"))
+        rename_keys.append((f"blocks.{i}.attn.qkv.bias", f"model.encoder.layers.{i}.attention.qkv.bias"))
+        rename_keys.append((f"blocks.{i}.attn.proj.weight", f"model.encoder.layers.{i}.attention.proj.weight"))
+        rename_keys.append((f"blocks.{i}.attn.proj.bias", f"model.encoder.layers.{i}.attention.proj.bias"))
+        rename_keys.append((f"blocks.{i}.attn.rel_pos_h", f"model.encoder.layers.{i}.attention.rel_pos_h"))
+        rename_keys.append((f"blocks.{i}.attn.rel_pos_w", f"model.encoder.layers.{i}.attention.rel_pos_w"))
+
+        rename_keys.append((f"blocks.{i}.mlp.fc1.weight", f"model.encoder.layers.{i}.mlp.lin1.weight"))
+        rename_keys.append((f"blocks.{i}.mlp.fc1.bias", f"model.encoder.layers.{i}.mlp.lin1.bias"))
+        rename_keys.append((f"blocks.{i}.mlp.fc2.weight", f"model.encoder.layers.{i}.mlp.lin2.weight"))
+        rename_keys.append((f"blocks.{i}.mlp.fc2.bias", f"model.encoder.layers.{i}.mlp.lin2.bias"))
+
+        rename_keys.append((f"blocks.{i}.norm1.weight", f"model.encoder.layers.{i}.layernorm_before.weight"))
+        rename_keys.append((f"blocks.{i}.norm1.bias", f"model.encoder.layers.{i}.layernorm_before.bias"))
+        rename_keys.append((f"blocks.{i}.norm2.weight", f"model.encoder.layers.{i}.layernorm_after.weight"))
+        rename_keys.append((f"blocks.{i}.norm2.bias", f"model.encoder.layers.{i}.layernorm_after.bias"))
+
+    # fmt: on
+
+    return rename_keys
+
+
+def rename_key(dct, old, new):
+    val = dct.pop(old)
+    dct[new] = val
+
+
+# We will verify our results on spongebob images
+def prepare_input():
+    image_input_url = (
+        "https://raw.githubusercontent.com/baaivision/Painter/main/SegGPT/SegGPT_inference/examples/hmbb_2.jpg"
+    )
+    image_prompt_url = (
+        "https://raw.githubusercontent.com/baaivision/Painter/main/SegGPT/SegGPT_inference/examples/hmbb_1.jpg"
+    )
+    mask_prompt_url = (
+        "https://raw.githubusercontent.com/baaivision/Painter/main/SegGPT/SegGPT_inference/examples/hmbb_1_target.png"
+    )
+
+    image_input = Image.open(requests.get(image_input_url, stream=True).raw)
+    image_prompt = Image.open(requests.get(image_prompt_url, stream=True).raw)
+    mask_prompt = Image.open(requests.get(mask_prompt_url, stream=True).raw)
+
+    return image_input, image_prompt, mask_prompt
+
+
+@torch.no_grad()
+def convert_seggpt_checkpoint(args):
+    model_name = args.model_name
+    pytorch_dump_folder_path = args.pytorch_dump_folder_path
+    verify_logits = args.verify_logits
+    push_to_hub = args.push_to_hub
+
+    # Define default GroundingDINO configuation
+    config = SegGptConfig()
+
+    # Load original checkpoint
+    checkpoint_url = "https://huggingface.co/BAAI/SegGpt/blob/main/seggpt_vit_large.pth"
+    original_state_dict = torch.hub.load_state_dict_from_url(checkpoint_url, map_location="cpu")["model"]
+
+    # # Rename keys
+    new_state_dict = original_state_dict.copy()
+    rename_keys = create_rename_keys(config)
+
+    for src, dest in rename_keys:
+        rename_key(new_state_dict, src, dest)
+
+    # Load HF model
+    model = SegGptForImageSegmentation(config)
+    model.eval()
+    missing_keys, unexpected_keys = model.load_state_dict(new_state_dict, strict=False)
+    print("Missing keys:", missing_keys)
+    print("Unexpected keys:", unexpected_keys)
+
+    input_img, prompt_img, prompt_mask = prepare_input()
+    image_processor = SegGptImageProcessor()
+    inputs = image_processor(images=input_img, prompt_images=prompt_img, prompt_masks=prompt_mask, return_tensors="pt")
+
+    expected_prompt_pixel_values = torch.tensor(
+        [
+            [[-0.6965, -0.6965, -0.6965], [-0.6965, -0.6965, -0.6965], [-0.6965, -0.6965, -0.6965]],
+            [[1.6583, 1.6583, 1.6583], [1.6583, 1.6583, 1.6583], [1.6583, 1.6583, 1.6583]],
+            [[2.3088, 2.3088, 2.3088], [2.3088, 2.3088, 2.3088], [2.3088, 2.3088, 2.3088]],
+        ]
+    )
+
+    expected_pixel_values = torch.tensor(
+        [
+            [[1.6324, 1.6153, 1.5810], [1.6153, 1.5982, 1.5810], [1.5810, 1.5639, 1.5639]],
+            [[1.2731, 1.2556, 1.2206], [1.2556, 1.2381, 1.2031], [1.2206, 1.2031, 1.1681]],
+            [[1.6465, 1.6465, 1.6465], [1.6465, 1.6465, 1.6465], [1.6291, 1.6291, 1.6291]],
+        ]
+    )
+
+    expected_prompt_masks = torch.tensor(
+        [
+            [[-2.1179, -2.1179, -2.1179], [-2.1179, -2.1179, -2.1179], [-2.1179, -2.1179, -2.1179]],
+            [[-2.0357, -2.0357, -2.0357], [-2.0357, -2.0357, -2.0357], [-2.0357, -2.0357, -2.0357]],
+            [[-1.8044, -1.8044, -1.8044], [-1.8044, -1.8044, -1.8044], [-1.8044, -1.8044, -1.8044]],
+        ]
+    )
+
+    assert torch.allclose(inputs.pixel_values[0, :, :3, :3], expected_pixel_values, atol=1e-4)
+    assert torch.allclose(inputs.prompt_pixel_values[0, :, :3, :3], expected_prompt_pixel_values, atol=1e-4)
+    assert torch.allclose(inputs.prompt_masks[0, :, :3, :3], expected_prompt_masks, atol=1e-4)
+
+    torch.manual_seed(2)
+    outputs = model(**inputs)
+    print(outputs)
+
+    if verify_logits:
+        expected_output = torch.tensor(
+            [
+                [[-2.1208, -2.1190, -2.1198], [-2.1237, -2.1228, -2.1227], [-2.1232, -2.1226, -2.1228]],
+                [[-2.0405, -2.0396, -2.0403], [-2.0434, -2.0434, -2.0433], [-2.0428, -2.0432, -2.0434]],
+                [[-1.8102, -1.8088, -1.8099], [-1.8131, -1.8126, -1.8129], [-1.8130, -1.8128, -1.8131]],
+            ]
+        )
+        assert torch.allclose(outputs.pred_masks[0, :, :3, :3], expected_output, atol=1e-4)
+        print("Looks good!")
+    else:
+        print("Converted without verifying logits")
+
+    if pytorch_dump_folder_path is not None:
+        print(f"Saving model and processor for {model_name} to {pytorch_dump_folder_path}")
+        model.save_pretrained(pytorch_dump_folder_path)
+        image_processor.save_pretrained(pytorch_dump_folder_path)
+
+    if push_to_hub:
+        print(f"Pushing model and processor for {model_name} to hub")
+        model.push_to_hub(f"EduardoPacheco/{model_name}")
+        image_processor.push_to_hub(f"EduardoPacheco/{model_name}")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--model_name",
+        default="seggpt-vit-large",
+        type=str,
+        choices=["seggpt-vit-large"],
+        help="Name of the SegGpt 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(
+        "--verify_logits",
+        action="store_false",
+        help="Whether or not to verify the logits against the original implementation.",
+    )
+    parser.add_argument(
+        "--push_to_hub", action="store_true", help="Whether or not to push the converted model to the 🤗 hub."
+    )
+
+    args = parser.parse_args()
+    convert_seggpt_checkpoint(args)