更换文档检测模型

paddle_detection/configs/group_detr/README.md

@@ -0,0 +1,49 @@
+# Group DETR: Fast DETR training with group-wise one-to-many assignment
+# Group DETR v2: Strong object detector with encoder-decoder pretraining
+
+## Introduction
+
+[Group DETR](https://arxiv.org/pdf/2207.13085.pdf) is an object detection model based on DETR. We reproduced the model of the paper.
+
+[Group DETR v2](https://arxiv.org/pdf/2211.03594.pdf) is a strong object detection model based on DINO and Group DETR. We reproduced the model of the paper.
+
+## Model Zoo
+
+| Backbone |      Model      | Epochs | Resolution |Box AP |                 Config                  |                                     Download                                     |
+|:------:|:---------------:|:------:|:------:|:---------------------------------------:|:--------------------------------------------------------------------------------:|:------:|
+| R-50 | dino_r50_4scale |   12   | (800, 1333) |  49.6  | [config](./group_dino_r50_4scale_1x_coco.yml) | [model](https://bj.bcebos.com/v1/paddledet/models/group_dino_r50_4scale_1x_coco.pdparams) |
+| Vit-huge | dino_vit_huge_4scale |   12   | (1184, 2000) | 63.3  | [config](./group_dino_vit_huge_4scale_1x_coco.yml) | [model](https://bj.bcebos.com/v1/paddledet/models/group_dino_vit_huge_4scale_1x_coco.pdparams) |
+
+**Notes:**
+
+- Group DETR is trained on COCO train2017 dataset and evaluated on val2017 results of `mAP(IoU=0.5:0.95)`.
+- Group DETRv2 requires a ViT-Huge encoder pre-trained and fine-tuned on ImageNet-1K in a self-supervised manner, a detector pre-trained on Object365, and finally it is fine-tuned on trainCOCO. Group DETRv2 is also evaluated on val2017 results of `mAP(IoU=0.5:0.95)`.
+- Group DETR and Group DETRv2 are both use 4GPU to train.
+
+GPU multi-card training
+```bash
+python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c configs/group_detr/group_dino_r50_4scale_1x_coco.yml --fleet --eval
+```
+
+```bash
+python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c configs/group_detr/group_dino_vit_huge_4scale_1x_coco.yml --fleet --eval
+```
+
+## Citations
+```
+@article{chen2022group,
+  title={Group DETR: Fast DETR training with group-wise one-to-many assignment},
+  author={Chen, Qiang and Chen, Xiaokang and Wang, Jian and Feng, Haocheng and Han, Junyu and Ding, Errui and Zeng, Gang and Wang, Jingdong},
+  journal={arXiv preprint arXiv:2207.13085},
+  volume={1},
+  number={2},
+  year={2022}
+}
+
+@article{chen2022group,
+  title={Group DETR v2: Strong object detector with encoder-decoder pretraining},
+  author={Chen, Qiang and Wang, Jian and Han, Chuchu and Zhang, Shan and Li, Zexian and Chen, Xiaokang and Chen, Jiahui and Wang, Xiaodi and Han, Shuming and Zhang, Gang and others},
+  journal={arXiv preprint arXiv:2211.03594},
+  year={2022}
+}
+```

paddle_detection/configs/group_detr/_base_/dino_2000_reader.yml

@@ -0,0 +1,44 @@
+worker_num: 2
+TrainReader:
+  sample_transforms:
+  - Decode: {}
+  - RandomFlip: {prob: 0.5}
+  - RandomSelect: { transforms1: [ RandomShortSideResize: { short_side_sizes: [480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832, 864, 896, 928, 960, 992, 1024, 1056, 1088, 1120, 1152, 1184], max_size: 2000 } ],
+                    transforms2: [
+                        RandomShortSideResize: { short_side_sizes: [400, 500, 600, 700, 800, 900] },
+                        RandomSizeCrop: { min_size: 384, max_size: 900 },
+                        RandomShortSideResize: { short_side_sizes: [480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832, 864, 896, 928, 960, 992, 1024, 1056, 1088, 1120, 1152, 1184], max_size: 2000 } ]
+  }
+  - NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
+  - NormalizeBox: {}
+  - BboxXYXY2XYWH: {}
+  - Permute: {}
+  batch_transforms:
+  - PadMaskBatch: {pad_to_stride: -1, return_pad_mask: true}
+  batch_size: 2
+  shuffle: true
+  drop_last: true
+  collate_batch: false
+  use_shared_memory: false
+
+
+EvalReader:
+  sample_transforms:
+  - Decode: {}
+  - Resize: {target_size: [1184, 2000], keep_ratio: True}
+  - NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
+  - Permute: {}
+  batch_size: 1
+  shuffle: false
+  drop_last: false
+
+
+TestReader:
+  sample_transforms:
+  - Decode: {}
+  - Resize: {target_size: [1184, 2000], keep_ratio: True}
+  - NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
+  - Permute: {}
+  batch_size: 1
+  shuffle: false
+  drop_last: false

paddle_detection/configs/group_detr/_base_/dino_reader.yml

@@ -0,0 +1,44 @@
+worker_num: 2
+TrainReader:
+  sample_transforms:
+  - Decode: {}
+  - RandomFlip: {prob: 0.5}
+  - RandomSelect: { transforms1: [ RandomShortSideResize: { short_side_sizes: [ 480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800 ], max_size: 1333 } ],
+                    transforms2: [
+                        RandomShortSideResize: { short_side_sizes: [ 400, 500, 600 ] },
+                        RandomSizeCrop: { min_size: 384, max_size: 600 },
+                        RandomShortSideResize: { short_side_sizes: [ 480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800 ], max_size: 1333 } ]
+  }
+  - NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
+  - NormalizeBox: {}
+  - BboxXYXY2XYWH: {}
+  - Permute: {}
+  batch_transforms:
+  - PadMaskBatch: {pad_to_stride: -1, return_pad_mask: true}
+  batch_size: 2
+  shuffle: true
+  drop_last: true
+  collate_batch: false
+  use_shared_memory: false
+
+
+EvalReader:
+  sample_transforms:
+  - Decode: {}
+  - Resize: {target_size: [800, 1333], keep_ratio: True}
+  - NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
+  - Permute: {}
+  batch_size: 1
+  shuffle: false
+  drop_last: false
+
+
+TestReader:
+  sample_transforms:
+  - Decode: {}
+  - Resize: {target_size: [800, 1333], keep_ratio: True}
+  - NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}
+  - Permute: {}
+  batch_size: 1
+  shuffle: false
+  drop_last: false

paddle_detection/configs/group_detr/_base_/group_dino_r50.yml

@@ -0,0 +1,53 @@
+architecture: DETR
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_cos_pretrained.pdparams
+hidden_dim: 256
+use_focal_loss: True
+
+
+DETR:
+  backbone: ResNet
+  transformer: GroupDINOTransformer
+  detr_head: DINOHead
+  post_process: DETRPostProcess
+
+ResNet:
+  # index 0 stands for res2
+  depth: 50
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [1, 2, 3]
+  lr_mult_list: [0.0, 0.1, 0.1, 0.1]
+  num_stages: 4
+
+GroupDINOTransformer:
+  num_queries: 900
+  position_embed_type: sine
+  num_levels: 4
+  nhead: 8
+  num_encoder_layers: 6
+  num_decoder_layers: 6
+  dim_feedforward: 2048
+  dropout: 0.0
+  activation: relu
+  pe_temperature: 20
+  pe_offset: 0.0
+  num_denoising: 100
+  label_noise_ratio: 0.5
+  box_noise_scale: 1.0
+  learnt_init_query: True
+  dual_queries: True
+  dual_groups: 10
+
+DINOHead:
+  loss:
+    name: DINOLoss
+    loss_coeff: {class: 1, bbox: 5, giou: 2}
+    aux_loss: True
+    matcher:
+      name: HungarianMatcher
+      matcher_coeff: {class: 2, bbox: 5, giou: 2}
+
+DETRPostProcess:
+  num_top_queries: 300
+  dual_queries: True
+  dual_groups: 10

paddle_detection/configs/group_detr/_base_/group_dino_vit_huge.yml

@@ -0,0 +1,68 @@
+architecture: DETR
+pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/vit_huge_mae_patch14_dec512d8b_pretrained.pdparams
+hidden_dim: 256
+use_focal_loss: True
+
+DETR:
+  backbone: VisionTransformer2D
+  neck: SimpleFeaturePyramid
+  transformer: GroupDINOTransformer
+  detr_head: DINOHead
+  post_process: DETRPostProcess
+
+VisionTransformer2D:
+  patch_size: 16
+  embed_dim: 1280
+  depth: 32
+  num_heads: 16
+  mlp_ratio: 4
+  attn_bias: True
+  drop_rate: 0.0
+  drop_path_rate: 0.1
+  lr_decay_rate: 0.7
+  global_attn_indexes: [7, 15, 23, 31]
+  use_abs_pos: False
+  use_rel_pos: True
+  rel_pos_zero_init: True
+  window_size: 14
+  out_indices: [ 31, ]
+
+SimpleFeaturePyramid:
+  out_channels: 256
+  num_levels: 4
+
+GroupDINOTransformer:
+  num_queries: 900
+  position_embed_type: sine
+  pe_temperature: 20
+  pe_offset: 0.0
+  num_levels: 4
+  nhead: 8
+  num_encoder_layers: 6
+  num_decoder_layers: 6
+  dim_feedforward: 2048
+  use_input_proj: False
+  dropout: 0.0
+  activation: relu
+  num_denoising: 100
+  label_noise_ratio: 0.5
+  box_noise_scale: 1.0
+  learnt_init_query: True
+  dual_queries: True
+  dual_groups: 10
+
+
+DINOHead:
+  loss:
+    name: DINOLoss
+    loss_coeff: {class: 1, bbox: 5, giou: 2}
+    aux_loss: True
+    matcher:
+      name: HungarianMatcher
+      matcher_coeff: {class: 2, bbox: 5, giou: 2}
+
+
+DETRPostProcess:
+  num_top_queries: 300
+  dual_queries: True
+  dual_groups: 10

paddle_detection/configs/group_detr/_base_/optimizer_1x.yml

@@ -0,0 +1,16 @@
+epoch: 12
+
+LearningRate:
+  base_lr: 0.0001
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [11]
+    use_warmup: false
+
+OptimizerBuilder:
+  clip_grad_by_norm: 0.1
+  regularizer: false
+  optimizer:
+    type: AdamW
+    weight_decay: 0.0001

paddle_detection/configs/group_detr/group_dino_r50_4scale_1x_coco.yml

@@ -0,0 +1,11 @@
+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_1x.yml',
+  '_base_/group_dino_r50.yml',
+  '_base_/dino_reader.yml',
+]
+
+weights: output/group_dino_r50_4scale_1x_coco/model_final
+find_unused_parameters: True
+log_iter: 100

paddle_detection/configs/group_detr/group_dino_vit_huge_4scale_1x_coco.yml

@@ -0,0 +1,11 @@
+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '_base_/optimizer_1x.yml',
+  '_base_/group_dino_vit_huge.yml',
+  '_base_/dino_2000_reader.yml',
+]
+
+weights: output/group_dino_vit_huge_4scale_1x_coco/model_final
+find_unused_parameters: True
+log_iter: 100