siton bug

configs/rtdetr/_base_/rtdetr_reader.yml

+worker_num: 4
+TrainReader:
+  sample_transforms:
+    - Decode: {}
+    - RandomDistort: {prob: 0.8}
+    - RandomExpand: {fill_value: [123.675, 116.28, 103.53]}
+    - RandomCrop: {prob: 0.8}
+    - RandomFlip: {}
+  batch_transforms:
+    - BatchRandomResize: {target_size: [480, 512, 544, 576, 608, 640, 640, 640, 672, 704, 736, 768, 800], random_size: True, random_interp: True, keep_ratio: False}
+    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
+    - NormalizeBox: {}
+    - BboxXYXY2XYWH: {}
+    - Permute: {}
+  batch_size: 4
+  shuffle: true
+  drop_last: true
+  collate_batch: false
+  use_shared_memory: true
+
+
+EvalReader:
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}
+    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
+    - Permute: {}
+  batch_size: 4
+
+
+TestReader:
+  inputs_def:
+    image_shape: [3, 640, 640]
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}
+    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
+    - Permute: {}
+  batch_size: 1

configs/rtdetr/rtdetr_focalnet_L_384_3x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_focalnet_L_384_3x_coco/model_final
+find_unused_parameters: True
+log_iter: 100
+snapshot_epoch: 2
+
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/focalnet_large_fl4_pretrained_on_o365.pdparams
+DETR:
+  backbone: FocalNet
+  neck: HybridEncoder
+  transformer: RTDETRTransformer
+  detr_head: DINOHead
+  post_process: DETRPostProcess
+
+FocalNet:
+  arch: 'focalnet_L_384_22k_fl4'
+  out_indices: [1, 2, 3]
+
+HybridEncoder:
+  hidden_dim: 256
+  use_encoder_idx: [2]
+  num_encoder_layers: 6 #
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 256
+    nhead: 8
+    dim_feedforward: 2048
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 1.0
+
+
+RTDETRTransformer:
+  num_queries: 300
+  position_embed_type: sine
+  feat_strides: [8, 16, 32]
+  num_levels: 3
+  nhead: 8
+  num_decoder_layers: 6
+  dim_feedforward: 2048 #
+  dropout: 0.0
+  activation: relu
+  num_denoising: 100
+  label_noise_ratio: 0.5
+  box_noise_scale: 1.0
+  learnt_init_query: False
+  query_pos_head_inv_sig: True #
+
+DINOHead:
+  loss:
+    name: DINOLoss
+    loss_coeff: {class: 1, bbox: 5, giou: 2}
+    aux_loss: True
+    use_vfl: True
+    matcher:
+      name: HungarianMatcher
+      matcher_coeff: {class: 2, bbox: 5, giou: 2}
+
+DETRPostProcess:
+  num_top_queries: 300
+
+
+epoch: 36
+LearningRate:
+  base_lr: 0.0001
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [36]
+    use_warmup: false
+
+OptimizerBuilder:
+  clip_grad_by_norm: 0.1
+  regularizer: false
+  optimizer:
+    type: AdamW
+    weight_decay: 0.0001
+    param_groups:
+      - params: ['absolute_pos_embed', 'relative_position_bias_table', 'norm']
+        weight_decay: 0.0

configs/rtdetr/rtdetr_hgnetv2_l_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_hgnetv2_l_6x_coco/model_final
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/PPHGNetV2_L_ssld_pretrained.pdparams
+find_unused_parameters: True
+log_iter: 200
+
+
+DETR:
+  backbone: PPHGNetV2
+
+PPHGNetV2:
+  arch: 'L'
+  return_idx: [1, 2, 3]
+  freeze_stem_only: True
+  freeze_at: 0
+  freeze_norm: True
+  lr_mult_list: [0., 0.05, 0.05, 0.05, 0.05]

configs/rtdetr/rtdetr_hgnetv2_x_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_hgnetv2_l_6x_coco/model_final
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/PPHGNetV2_X_ssld_pretrained.pdparams
+find_unused_parameters: True
+log_iter: 200
+
+
+
+DETR:
+  backbone: PPHGNetV2
+
+
+PPHGNetV2:
+  arch: 'X'
+  return_idx: [1, 2, 3]
+  freeze_stem_only: True
+  freeze_at: 0
+  freeze_norm: True
+  lr_mult_list: [0., 0.01, 0.01, 0.01, 0.01]
+
+
+HybridEncoder:
+  hidden_dim: 384
+  use_encoder_idx: [2]
+  num_encoder_layers: 1
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 384
+    nhead: 8
+    dim_feedforward: 2048
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 1.0

configs/rtdetr/rtdetr_r101vd_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_r101vd_6x_coco/model_final
+find_unused_parameters: True
+log_iter: 200
+
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet101_vd_ssld_pretrained.pdparams
+
+ResNet:
+  # index 0 stands for res2
+  depth: 101
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [1, 2, 3]
+  lr_mult_list: [0.01, 0.01, 0.01, 0.01]
+  num_stages: 4
+  freeze_stem_only: True
+
+HybridEncoder:
+  hidden_dim: 384
+  use_encoder_idx: [2]
+  num_encoder_layers: 1
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 384
+    nhead: 8
+    dim_feedforward: 2048
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 1.0

configs/rtdetr/rtdetr_r18vd_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_r18_6x_coco/model_final
+find_unused_parameters: True
+log_iter: 200
+
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet18_vd_pretrained.pdparams
+ResNet:
+  depth: 18
+  variant: d
+  return_idx: [1, 2, 3]
+  freeze_at: -1
+  freeze_norm: false
+  norm_decay: 0.
+
+HybridEncoder:
+  hidden_dim: 256
+  use_encoder_idx: [2]
+  num_encoder_layers: 1
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 256
+    nhead: 8
+    dim_feedforward: 1024
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 0.5
+  depth_mult: 1.0
+
+RTDETRTransformer:
+  eval_idx: -1
+  num_decoder_layers: 3

configs/rtdetr/rtdetr_r34vd_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_r34vd_6x_coco/model_final
+find_unused_parameters: True
+log_iter: 200
+
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/ResNet34_vd_pretrained.pdparams
+ResNet:
+  depth: 34
+  variant: d
+  return_idx: [1, 2, 3]
+  freeze_at: -1
+  freeze_norm: false
+  norm_decay: 0.
+
+HybridEncoder:
+  hidden_dim: 256
+  use_encoder_idx: [2]
+  num_encoder_layers: 1
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 256
+    nhead: 8
+    dim_feedforward: 1024
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 0.5
+  depth_mult: 1.0
+
+RTDETRTransformer:
+  eval_idx: -1
+  num_decoder_layers: 4

configs/rtdetr/rtdetr_r50vd_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_r50vd_6x_coco/model_final
+find_unused_parameters: True
+log_iter: 200

configs/rtdetr/rtdetr_r50vd_m_6x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_r50vd_m_6x_coco/model_final
+find_unused_parameters: True
+log_iter: 200
+
+HybridEncoder:
+  hidden_dim: 256
+  use_encoder_idx: [2]
+  num_encoder_layers: 1
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 256
+    nhead: 8
+    dim_feedforward: 1024
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 0.5
+  depth_mult: 1.0
+
+RTDETRTransformer:
+  eval_idx: 2 # use 3th decoder layer to eval

configs/rtdetr/rtdetr_swin_L_384_3x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_6x.yml',
+  '_base_/rtdetr_r50vd.yml',
+  '_base_/rtdetr_reader.yml',
+]
+
+weights: output/rtdetr_swin_L_384_3x_coco/model_final
+find_unused_parameters: True
+log_iter: 100
+snapshot_epoch: 2
+
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/dino_swin_large_384_4scale_3x_coco.pdparams
+DETR:
+  backbone: SwinTransformer
+  neck: HybridEncoder
+  transformer: RTDETRTransformer
+  detr_head: DINOHead
+  post_process: DETRPostProcess
+
+
+SwinTransformer:
+  arch: 'swin_L_384' # ['swin_T_224', 'swin_S_224', 'swin_B_224', 'swin_L_224', 'swin_B_384', 'swin_L_384']
+  ape: false
+  drop_path_rate: 0.2
+  patch_norm: true
+  out_indices: [1, 2, 3]
+
+HybridEncoder:
+  hidden_dim: 256
+  use_encoder_idx: [2]
+  num_encoder_layers: 6 #
+  encoder_layer:
+    name: TransformerLayer
+    d_model: 256
+    nhead: 8
+    dim_feedforward: 2048 #
+    dropout: 0.
+    activation: 'gelu'
+  expansion: 1.0
+
+RTDETRTransformer:
+  num_queries: 300
+  position_embed_type: sine
+  feat_strides: [8, 16, 32]
+  num_levels: 3
+  nhead: 8
+  num_decoder_layers: 6
+  dim_feedforward: 2048 #
+  dropout: 0.0
+  activation: relu
+  num_denoising: 100
+  label_noise_ratio: 0.5
+  box_noise_scale: 1.0
+  learnt_init_query: False
+
+DINOHead:
+  loss:
+    name: DINOLoss
+    loss_coeff: {class: 1, bbox: 5, giou: 2}
+    aux_loss: True
+    use_vfl: True
+    matcher:
+      name: HungarianMatcher
+      matcher_coeff: {class: 2, bbox: 5, giou: 2}
+
+DETRPostProcess:
+  num_top_queries: 300
+
+
+epoch: 36
+LearningRate:
+  base_lr: 0.0001
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [36]
+    use_warmup: false
+
+OptimizerBuilder:
+  clip_grad_by_norm: 0.1
+  regularizer: false
+  optimizer:
+    type: AdamW
+    weight_decay: 0.0001
+    param_groups:
+      - params: ['absolute_pos_embed', 'relative_position_bias_table', 'norm']
+        weight_decay: 0.0

configs/rtmdet/README.md

+# RTMDet
+
+## 内容
+- [模型库](#模型库)
+- [使用说明](#使用说明)
+- [速度测试](#速度测试)
+- [引用](#引用)
+
+## 模型库
+
+### 基础检测模型
+
+| 网络网络        | 输入尺寸   | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) |   mAP  |   AP50  | Params(M) | FLOPs(G) |  下载链接       | 配置文件 |
+| :------------- | :------- | :-------: | :------: | :---------: | :-----: |:-----: | :-----: |:-----: | :-------------: | :-----: |
+| *RTMDet-t       |  640     |    32      |   300e    |    2.8   |  40.9 | 57.9 |  4.90  | 16.21 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_t_300e_coco.pdparams) | [配置文件](./rtmdet_t_300e_coco.yml) |
+| *RTMDet-s       |  640     |    32      |   300e    |    3.3   |  44.5 | 62.0 |  8.89  | 29.71 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams) | [配置文件](./rtmdet_s_300e_coco.yml) |
+| *RTMDet-m       |  640     |    32      |   300e    |    6.4   |  49.1 | 66.8 |  24.71  | 78.47 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_m_300e_coco.pdparams) | [配置文件](./rtmdet_m_300e_coco.yml) |
+| *RTMDet-l       |  640     |    32      |   300e    |    10.2  |  51.2 | 68.8 |  52.31  | 160.32 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_l_300e_coco.pdparams) | [配置文件](./rtmdet_l_300e_coco.yml) |
+| *RTMDet-x       |  640     |    32      |   300e    |    18.0  |  52.6 | 70.4 |  94.86  | 283.12 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_x_300e_coco.pdparams) | [配置文件](./rtmdet_x_300e_coco.yml) |
+
+### 实例分割模型
+
+| 网络网络        | 输入尺寸   | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) |  box AP  |  mask AP  | Params(M) | FLOPs(G) |  下载链接       | 配置文件 |
+| :------------- | :------- | :-------: | :------: | :---------: | :-----: |:-----: | :-----: |:-----: | :-------------: | :-----: |
+| *RTMDet-t       |  640     |    32      |   300e    |    -   |  40.5 | - |  5.6  | 11.8 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_ins_t_300e_coco.pdparams) | [配置文件](./rtmdet_ins_t_300e_coco.yml) |
+| *RTMDet-s       |  640     |    32      |   300e    |    -   |  44.0 | - |  10.18  | 21.5 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_ins_s_300e_coco.pdparams) | [配置文件](./rtmdet_ins_s_300e_coco.yml) |
+| *RTMDet-m       |  640     |    32      |   300e    |    -   |  48.8 | - |  27.58  | 54.13 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_ins_m_300e_coco.pdparams) | [配置文件](./rtmdet_ins_m_300e_coco.yml) |
+| *RTMDet-l       |  640     |    32      |   300e    |    -  |  51.2 | - |  57.37  | 106.56 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_ins_l_300e_coco.pdparams) | [配置文件](./rtmdet_ins_l_300e_coco.yml) |
+| *RTMDet-x       |  640     |    32      |   300e    |    -  |  52.4 | - |  102.7  | 182.7 |[下载链接](https://paddledet.bj.bcebos.com/models/rtmdet_ins_x_300e_coco.pdparams) | [配置文件](./rtmdet_ins_x_300e_coco.yml) |
+
+**注意:**
+  - RTMDet模型暂未支持完全训练，mAP为部署权重在COCO val2017上的`mAP(IoU=0.5:0.95)`结果，且评估未使用`multi_label`等trick；
+  - RTMDet t s模型训练使用了ImageNet预训练权重，其余 m l x 模型未使用；
+  - RTMDet模型Params(M)和FLOPs(G)均为训练时所测；
+  - RTMDet模型训练过程中默认使用8 GPUs进行混合精度训练，默认每卡batch_size=32；
+  - RTMDet模型原文测速是在`NVIDIA 3090 GPU`上，暂未提供T4、V100上的测速数据；
+  - 模型推理耗时(ms)为TensorRT-FP16下测试的耗时，不包含数据预处理和模型输出后处理(NMS)的耗时。测试采用单卡Tesla T4 GPU，batch size=1，测试环境为**paddlepaddle-2.3.2**, **CUDA 11.2**, **CUDNN 8.2**, **GCC-8.2**, **TensorRT 8.0.3.4**，具体请参考[速度测试](#速度测试)。
+  - 如果你设置了`--run_benchmark=True`, 你首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
+
+
+### 部署模型
+
+| 网络模型     | 输入尺寸 | 导出后的权重(w/o NMS) | ONNX(w/o NMS)  |
+| :-------- | :--------: | :---------------------: | :----------------: |
+| RTMDet-t |  640   | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_t_300e_coco_w_nms.zip) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_t_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_t_300e_coco_w_nms.onnx) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_t_300e_coco_wo_nms.onnx) |
+| RTMDet-s |  640   | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_s_300e_coco_w_nms.zip) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_s_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_s_300e_coco_w_nms.onnx) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_s_300e_coco_wo_nms.onnx) |
+| RTMDet-m |  640   | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_m_300e_coco_w_nms.zip) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_m_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_m_300e_coco_w_nms.onnx) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_m_300e_coco_wo_nms.onnx) |
+| RTMDet-l |  640   | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_l_300e_coco_w_nms.zip) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_l_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_l_300e_coco_w_nms.onnx) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_l_300e_coco_wo_nms.onnx) |
+| RTMDet-x |  640   | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_x_300e_coco_w_nms.zip) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_x_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_x_300e_coco_w_nms.onnx) | [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/rtmdet/rtmdet_x_300e_coco_wo_nms.onnx) |
+
+
+## 使用教程
+
+### 0. **一键运行全流程**
+
+将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。
+
+```bash
+model_name=rtmdet # 可修改，如 ppyoloe
+job_name=rtmdet_s_300e_coco # 可修改，如 ppyoloe_plus_crn_s_80e_coco
+
+config=configs/${model_name}/${job_name}.yml
+log_dir=log_dir/${job_name}
+# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
+weights=output/${job_name}/model_final.pdparams
+
+# 1.训练（单卡/多卡），加 --eval 表示边训边评估，加 --amp 表示混合精度训练
+# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp
+python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp
+
+# 2.评估，加 --classwise 表示输出每一类mAP
+CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise
+
+# 3.预测 (单张图/图片文件夹）
+CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5
+# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5
+
+# 4.导出模型，以下3种模式选一种
+## 普通导出，加trt表示用于trt加速，对NMS和silu激活函数提速明显
+CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True
+
+## exclude_post_process去除后处理导出，返回和YOLOv5导出ONNX时相同格式的concat后的1个Tensor，是未缩放回原图的坐标+分类置信度
+# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True
+
+## exclude_nms去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度
+# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True
+
+# 5.部署预测，注意不能使用 去除后处理 或 去除NMS 导出后的模型去预测
+CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU
+
+# 6.部署测速，加 “--run_mode=trt_fp16” 表示在TensorRT FP16模式下测速，注意如需用到 trt_fp16 则必须为加 trt=True 导出的模型
+CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16
+
+# 7.onnx导出，一般结合 exclude_post_process去除后处理导出的模型
+paddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx
+
+# 8.onnx trt测速
+/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16
+/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32
+```
+
+### 1. 训练
+执行以下指令使用混合精度训练rtmdet
+```bash
+python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/rtmdet/rtmdet_s_300e_coco.yml --amp --eval
+```
+**注意:**
+- `--amp`表示开启混合精度训练以避免显存溢出，`--eval`表示边训边验证。
+
+### 2. 评估
+执行以下命令在单个GPU上评估COCO val2017数据集
+```bash
+CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams
+```
+
+### 3. 推理
+使用以下命令在单张GPU上预测图片，使用`--infer_img`推理单张图片以及使用`--infer_dir`推理文件中的所有图片。
+```bash
+# 推理单张图片
+CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams --infer_img=demo/000000014439_640x640.jpg
+
+# 推理文件中的所有图片
+CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams --infer_dir=demo
+```
+
+### 4.导出模型
+在GPU上推理部署或benchmark测速等需要通过`tools/export_model.py`导出模型。
+
+当你**使用Paddle Inference但不使用TensorRT**时，运行以下的命令导出模型
+
+```bash
+python tools/export_model.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams
+```
+
+当你**使用Paddle Inference且使用TensorRT**时，需要指定`-o trt=True`来导出模型。
+
+```bash
+python tools/export_model.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams trt=True
+```
+
+如果你想将rtmdet模型导出为**ONNX格式**，参考
+[PaddleDetection模型导出为ONNX格式教程](../../deploy/EXPORT_ONNX_MODEL.md)，运行以下命令：
+
+```bash
+
+# 导出推理模型
+python tools/export_model.py -c configs/rtmdet/rtmdet_s_300e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams
+
+# 安装paddle2onnx
+pip install paddle2onnx
+
+# 转换成onnx格式
+paddle2onnx --model_dir output_inference/rtmdet_s_300e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 11 --save_file rtmdet_s_300e_coco.onnx
+```
+
+**注意：** ONNX模型目前只支持batch_size=1
+
+
+### 5.推理部署
+rtmdet可以使用以下方式进行部署：
+  - Paddle Inference [Python](../../deploy/python) & [C++](../../deploy/cpp)
+  - [Paddle-TensorRT](../../deploy/TENSOR_RT.md)
+  - [PaddleServing](https://github.com/PaddlePaddle/Serving)
+  - [PaddleSlim模型量化](../slim)
+
+运行以下命令导出模型
+
+```bash
+python tools/export_model.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams trt=True
+```
+
+**注意：**
+- trt=True表示**使用Paddle Inference且使用TensorRT**进行测速，速度会更快，默认不加即为False，表示**使用Paddle Inference但不使用TensorRT**进行测速。
+- 如果是使用Paddle Inference在TensorRT FP16模式下部署，需要参考[Paddle Inference文档](https://www.paddlepaddle.org.cn/inference/master/user_guides/download_lib.html#python)，下载并安装与你的CUDA, CUDNN和TensorRT相应的wheel包。
+
+#### 5.1.Python部署
+`deploy/python/infer.py`使用上述导出后的Paddle Inference模型用于推理和benchnark测速，如果设置了`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
+
+```bash
+# Python部署推理单张图片
+python deploy/python/infer.py --model_dir=output_inference/rtmdet_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu
+
+# 推理文件夹下的所有图片
+python deploy/python/infer.py --model_dir=output_inference/rtmdet_s_300e_coco --image_dir=demo/ --device=gpu
+```
+
+#### 5.2. C++部署
+`deploy/cpp/build/main`使用上述导出后的Paddle Inference模型用于C++推理部署, 首先按照[docs](../../deploy/cpp/docs)编译安装环境。
+```bash
+# C++部署推理单张图片
+./deploy/cpp/build/main --model_dir=output_inference/rtmdet_s_300e_coco/ --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=GPU --threshold=0.5 --output_dir=cpp_infer_output/rtmdet_s_300e_coco
+```
+
+
+## 速度测试
+
+为了公平起见，在[模型库](#模型库)中的速度测试结果均为不包含数据预处理和模型输出后处理(NMS)的数据(与[YOLOv4(AlexyAB)](https://github.com/AlexeyAB/darknet)测试方法一致)，需要在导出模型时指定`-o exclude_nms=True`。测速需设置`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
+
+**使用Paddle Inference但不使用TensorRT**进行测速，执行以下命令：
+
+```bash
+# 导出模型
+python tools/export_model.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams exclude_nms=True
+
+# 速度测试，使用run_benchmark=True
+python deploy/python/infer.py --model_dir=output_inference/rtmdet_s_300e_coco --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=gpu --run_benchmark=True
+```
+
+**使用Paddle Inference且使用TensorRT**进行测速，执行以下命令：
+
+```bash
+# 导出模型，使用trt=True
+python tools/export_model.py -c configs/rtmdet/rtmdet_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/rtmdet_s_300e_coco.pdparams exclude_nms=True trt=True
+
+# 速度测试，使用run_benchmark=True
+python deploy/python/infer.py --model_dir=output_inference/rtmdet_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True
+
+# tensorRT-FP32测速
+python deploy/python/infer.py --model_dir=output_inference/rtmdet_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp32
+
+# tensorRT-FP16测速
+python deploy/python/infer.py --model_dir=output_inference/rtmdet_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True  --run_mode=trt_fp16
+```
+**注意:**
+- 导出模型时指定`-o exclude_nms=True`仅作为测速时用，这样导出的模型其推理部署预测的结果不是最终检出框的结果。
+- [模型库](#模型库)中的速度测试结果为tensorRT-FP16测速后的最快速度，为不包含数据预处理和模型输出后处理(NMS)的耗时。
+
+
+## 引用
+```
+@misc{lyu2022rtmdet,
+      title={RTMDet: An Empirical Study of Designing Real-Time Object Detectors},
+      author={Chengqi Lyu and Wenwei Zhang and Haian Huang and Yue Zhou and Yudong Wang and Yanyi Liu and Shilong Zhang and Kai Chen},
+      year={2022},
+      eprint={2212.07784},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```

configs/rtmdet/_base_/optimizer_300e.yml

+# set the same as YOLOX scheduler temporarily
+epoch: 300
+
+LearningRate:
+  base_lr: 0.0004
+  schedulers:
+    - !CosineDecay
+      max_epochs: 300
+    - !LinearWarmup
+      start_factor: 0.00001
+      steps: 1000
+
+OptimizerBuilder:
+  regularizer: false
+  optimizer:
+    type: AdamW
+    weight_decay: 0.05

configs/rtmdet/_base_/rtmdet_cspnext.yml

+architecture: RTMDet
+norm_type: sync_bn
+use_ema: True
+ema_decay: 0.9998
+ema_decay_type: "exponential"
+act: silu
+find_unused_parameters: True
+
+depth_mult: 1.0
+width_mult: 1.0
+
+RTMDet:
+  backbone: CSPNeXt
+  neck: CSPNeXtPAFPN
+  head: RTMDetHead
+  post_process: ~
+
+CSPNeXt:
+  arch: "P5"
+  return_idx: [2, 3, 4]
+
+# use default config
+# CSPNeXtPAFPN:
+
+RTMDetHead:
+  exp_on_reg: False
+  fpn_strides: [8, 16, 32]
+  grid_cell_offset: 0
+  nms:
+    name: MultiClassNMS
+    nms_top_k: 1000
+    keep_top_k: 100
+    score_threshold: 0.05
+    nms_threshold: 0.6

configs/rtmdet/_base_/rtmdet_ins_cspnext.yml

+architecture: RTMDet
+norm_type: sync_bn
+use_ema: True
+ema_decay: 0.9998
+ema_decay_type: "exponential"
+act: silu
+find_unused_parameters: True
+with_mask: True
+
+depth_mult: 1.0
+width_mult: 1.0
+
+RTMDet:
+  backbone: CSPNeXt
+  neck: CSPNeXtPAFPN
+  head: RTMDetInsHead
+  with_mask: True
+  post_process: ~
+
+CSPNeXt:
+  arch: "P5"
+  return_idx: [2, 3, 4]
+
+# use default config
+# CSPNeXtPAFPN:
+
+RTMDetInsHead:
+  exp_on_reg: False
+  fpn_strides: [8, 16, 32]
+  grid_cell_offset: 0
+  nms:
+    name: MultiClassNMS
+    nms_top_k: 1000
+    keep_top_k: 100
+    score_threshold: 0.05
+    nms_threshold: 0.6
+    return_index: True

configs/rtmdet/_base_/rtmdet_reader.yml

+# TrainReader is temporarily set the same as YOLOX' TrainReader
+# EvalReader and TestReader is the final RTMDet reader
+
+worker_num: 4
+TrainReader:
+  sample_transforms:
+    - Decode: {}
+    - Mosaic:
+        prob: 1.0
+        input_dim: [640, 640]
+        degrees: [-10, 10]
+        scale: [0.1, 2.0]
+        shear: [-2, 2]
+        translate: [-0.1, 0.1]
+        enable_mixup: True
+        mixup_prob: 1.0
+        mixup_scale: [0.5, 1.5]
+    - AugmentHSV: {is_bgr: False, hgain: 5, sgain: 30, vgain: 30}
+    - PadResize: {target_size: 640}
+    - RandomFlip: {}
+  batch_transforms:
+    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
+    - Permute: {}
+    - PadGT: {}
+  batch_size: 32
+  shuffle: True
+  drop_last: True
+  use_shared_memory: True
+  collate_batch: True
+  mosaic_epoch: 280
+
+
+EvalReader:
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: [640, 640], keep_ratio: True, interp: 1}
+    - Pad: {size: [640, 640], fill_value: [114., 114., 114.]}
+    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
+    - Permute: {}
+  batch_size: 1
+
+
+TestReader:
+  inputs_def:
+    image_shape: [3, 640, 640]
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: [640, 640], keep_ratio: True, interp: 1}
+    - Pad: {size: [640, 640], fill_value: [114., 114., 114.]}
+    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
+    - Permute: {}
+  batch_size: 1
+  fuse_normalize: False

configs/rtmdet/rtmdet_ins_l_300e_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  './_base_/optimizer_300e.yml',
+  './_base_/rtmdet_ins_cspnext.yml',
+  './_base_/rtmdet_reader.yml',
+]
+depth_mult: 1.0
+width_mult: 1.0
+
+log_iter: 100
+snapshot_epoch: 10
+weights: output/rtmnet_ins_l_300e_coco/model_final

configs/rtmdet/rtmdet_ins_m_300e_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  './_base_/optimizer_300e.yml',
+  './_base_/rtmdet_ins_cspnext.yml',
+  './_base_/rtmdet_reader.yml',
+]
+depth_mult: 0.67
+width_mult: 0.75
+
+log_iter: 100
+snapshot_epoch: 10
+weights: output/rtmnet_ins_m_300e_coco/model_final

configs/rtmdet/rtmdet_ins_s_300e_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  './_base_/optimizer_300e.yml',
+  './_base_/rtmdet_ins_cspnext.yml',
+  './_base_/rtmdet_reader.yml',
+]
+depth_mult: 0.33
+width_mult: 0.50
+
+log_iter: 100
+snapshot_epoch: 10
+weights: output/rtmnet_ins_s_300e_coco/model_final
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/cspnext_s_pretrained.pdparams

configs/rtmdet/rtmdet_ins_t_300e_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  './_base_/optimizer_300e.yml',
+  './_base_/rtmdet_ins_cspnext.yml',
+  './_base_/rtmdet_reader.yml',
+]
+depth_mult: 0.167 # 0.33 in yolox-tiny
+width_mult: 0.375
+
+log_iter: 100
+snapshot_epoch: 10
+weights: output/rtmnet_ins_t_300e_coco/model_final
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/cspnext_t_pretrained.pdparams

configs/rtmdet/rtmdet_ins_x_300e_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  './_base_/optimizer_300e.yml',
+  './_base_/rtmdet_ins_cspnext.yml',
+  './_base_/rtmdet_reader.yml',
+]
+depth_mult: 1.33
+width_mult: 1.25
+
+log_iter: 100
+snapshot_epoch: 10
+weights: output/rtmnet_ins_x_300e_coco/model_final