[smalldet] update visdrone modelzoo (#6526)

* [smalldet visdrone] add modle zoo

* delete

* [smdet] add visdrone sliced config

* fix visdrone smalldet doc, test=document_fix

configs/smalldet/README.md

@@ -7,6 +7,7 @@
 |:---------|:---------------:|:---------------:|:---------------:|:------:|:-----------------------:|:-------------------:|:---------:| :-----: |
 |PP-YOLOE-l|   Xview  |  400 | 0.25 | 60 |  14.5 | 26.8 | [下载链接](https://bj.bcebos.com/v1/paddledet/models/ppyoloe_crn_l_xview_400_025.pdparams) | [配置文件](./ppyoloe_crn_l_80e_sliced_xview_400_025.yml) |
 |PP-YOLOE-l|   DOTA   |  500 | 0.25 | 15 |  46.8 |  72.6 | [下载链接](https://bj.bcebos.com/v1/paddledet/models/ppyoloe_crn_l_dota_500_025.pdparams) | [配置文件](./ppyoloe_crn_l_80e_sliced_DOTA_500_025.yml) |
+|PP-YOLOE-l| VisDrone |  500 | 0.25 | 10 |  29.7 |  48.5 | [下载链接](https://bj.bcebos.com/v1/paddledet/models/ppyoloe_crn_l_80e_sliced_visdrone_640_025.pdparams) | [配置文件](./ppyoloe_crn_l_80e_sliced_visdrone_640_025.yml) |
 
 
 **注意:**

configs/smalldet/_base_/visdrone_sliced_640_025_detection.yml

+metric: COCO
+num_classes: 10
+
+TrainDataset:
+  !COCODataSet
+    image_dir: train_images_640_025
+    anno_path: train_640_025.json
+    dataset_dir: dataset/visdrone_sliced
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
+
+EvalDataset:
+  !COCODataSet
+    image_dir: val_images_640_025
+    anno_path: val_640_025.json
+    dataset_dir: dataset/visdrone_sliced
+
+TestDataset:
+  !ImageFolder
+    anno_path: dataset/visdrone_sliced/val_640_025.json
+    dataset_dir: dataset/visdrone_sliced/val_images_640_025

configs/visdrone/ppyoloe_crn_s_80e_visdrone_use_checkpoint.yml → configs/smalldet/ppyoloe_crn_l_80e_sliced_visdrone_640_025.yml

 _BASE_: [
-  '../datasets/visdrone_detection.yml',
+  './_base_/visdrone_sliced_640_025_detection.yml',
   '../runtime.yml',
   '../ppyoloe/_base_/optimizer_300e.yml',
   '../ppyoloe/_base_/ppyoloe_crn.yml',
@@ -7,11 +7,11 @@ _BASE_: [
 ]
 log_iter: 100
 snapshot_epoch: 10
-weights: output/ppyoloe_crn_s_80e_visdrone_use_checkpoint/model_final
+weights: output/ppyoloe_crn_l_80e_sliced_visdrone_640_025/model_final
 
-pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams
-depth_mult: 0.33
-width_mult: 0.50
+pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams
+depth_mult: 1.0
+width_mult: 1.0
 
 TrainReader:
   batch_size: 8
@@ -26,13 +26,6 @@ LearningRate:
       start_factor: 0.
       epochs: 1
 
-CSPResNet:
-  use_checkpoint: True
-  use_alpha: True
-
-# when use_checkpoint
-use_fused_allreduce_gradients: True
-
 PPYOLOEHead:
   static_assigner_epoch: -1
   nms:

configs/visdrone/README.md

-简体中文 | [English](README.md)
-
 # VisDrone-DET 检测模型
 
 PaddleDetection团队提供了针对VisDrone-DET小目标数航拍场景的基于PP-YOLOE的检测模型，用户可以下载模型进行使用。整理后的COCO格式VisDrone-DET数据集[下载链接](https://bj.bcebos.com/v1/paddledet/data/smalldet/visdrone.zip)，检测其中的10类，包括 `pedestrian(1), people(2), bicycle(3), car(4), van(5), truck(6), tricycle(7), awning-tricycle(8), bus(9), motor(10)`，原始数据集[下载链接](https://github.com/VisDrone/VisDrone-Dataset)。
 
+**注意:**
+- VisDrone-DET数据集包括train集6471张，val集548张，test_dev集1610张，test-challenge集1580张(未开放检测框标注)，前三者均有开放检测框标注。
+- 模型均只使用train集训练，在val集和test_dev集上验证精度，test_dev集图片数较多，精度参考性较高。
+
+
+## 原图训练：
+
 |    模型   | COCOAPI mAP<sup>val<br>0.5:0.95 | COCOAPI mAP<sup>val<br>0.5 | COCOAPI mAP<sup>test_dev<br>0.5:0.95 | COCOAPI mAP<sup>test_dev<br>0.5 | MatlabAPI mAP<sup>test_dev<br>0.5:0.95 | MatlabAPI mAP<sup>test_dev<br>0.5 | 下载  | 配置文件 |
 |:---------|:------:|:------:| :----: | :------:| :------: | :------:| :----: | :------:|
 |PP-YOLOE-s|  23.5  |  39.9  |  19.4  |  33.6   |  23.68   |  40.66  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_80e_visdrone.pdparams) | [配置文件](./ppyoloe_crn_s_80e_visdrone.yml) |
+|PP-YOLOE-P2-Alpha-s|    24.4  |  41.6  |  20.1  |  34.7  |  24.55   |  42.19  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_p2_alpha_80e_visdrone.pdparams) | [配置文件](./ppyoloe_crn_s_p2_alpha_80e_visdrone.yml) |
 |PP-YOLOE-l|  29.2  |  47.3  |  23.5  |  39.1   |  28.00   |  46.20  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_80e_visdrone.pdparams) | [配置文件](./ppyoloe_crn_l_80e_visdrone.yml) |
-|PP-YOLOE-P2-l|  30.0  |  49.2  |  24.1  |  40.9   |  28.47   |  48.16  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_p2_crn_l_80e_visdrone.pdparams) | [配置文件](./ppyoloe_p2_crn_l_80e_visdrone.yml) |
-|PP-YOLOE-l largesize|  40.3  |  63.5  |  31.3  |  51.8   |  36.13   |  59.96  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_80e_visdrone_largesize.pdparams) | [配置文件](./ppyoloe_crn_l_80e_visdrone_largesize.yml) |
+|PP-YOLOE-P2-Alpha-l|  30.1  |  48.9  |  24.3  |  40.8   |  28.47   |  48.16  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_p2_alpha_80e_visdrone.pdparams) | [配置文件](./ppyoloe_crn_l_p2_alpha_80e_visdrone.yml) |
+|PP-YOLOE-Alpha-largesize-l|  41.9  |  65.0 |  32.3  |  53.0   |  37.13   |  61.15  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_alpha_largesize_80e_visdrone.pdparams) | [配置文件](./ppyoloe_crn_l_alpha_largesize_80e_visdrone.yml) |
+|PP-YOLOE-P2-Alpha-largesize-l|  41.3  |  64.5  |  32.4  |  53.1   |  37.49   |  51.54  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_p2_alpha_largesize_80e_visdrone.pdparams) | [配置文件](./ppyoloe_crn_l_p2_alpha_largesize_80e_visdrone.yml) |
+
+## 切图训练：
+
+|    模型   | COCOAPI mAP<sup>val<br>0.5:0.95 | COCOAPI mAP<sup>val<br>0.5 | COCOAPI mAP<sup>test_dev<br>0.5:0.95 | COCOAPI mAP<sup>test_dev<br>0.5 | MatlabAPI mAP<sup>test_dev<br>0.5:0.95 | MatlabAPI mAP<sup>test_dev<br>0.5 | 下载  | 配置文件 |
+|:---------|:------:|:------:| :----: | :------:| :------: | :------:| :----: | :------:|
+|PP-YOLOE-l|  29.7  |  48.5  |  23.3  |  39.9   |  -   |   -  | [下载链接](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_80e_sliced_visdrone_640_025.pdparams) | [配置文件](../smalldet/ppyoloe_crn_l_80e_sliced_visdrone_640_025.yml) |
+
 
 **注意:**
 - PP-YOLOE模型训练过程中使用8 GPUs进行混合精度训练，如果**GPU卡数**或者**batch size**发生了改变，你需要按照公式 **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)** 调整学习率。
 - 具体使用教程请参考[ppyoloe](../ppyoloe#getting-start)。
-- PP-YOLOE-P2是指增加P2层(1/4下采样层)的特征，共输出4个PPYOLOEHead。
-- largesize是指使用以1600尺度为基础的多尺度训练和1920尺度预测，相应的训练batch_size也减小，以速度来换取高精度。
+- P2表示增加P2层(1/4下采样层)的特征，共输出4个PPYOLOEHead。
+- Alpha表示对CSPResNet骨干网络增加可一个学习权重参数Alpha参与训练。
+- largesize表示使用以1600尺度为基础的多尺度训练和1920尺度预测，相应的训练batch_size也减小，以速度来换取高精度。
 - MatlabAPI测试是使用官网评测工具[VisDrone2018-DET-toolkit](https://github.com/VisDrone/VisDrone2018-DET-toolkit)。
+- 切图训练模型的配置文件及训练相关流程请参照[smalldet](../smalldet)。
 
 
 ## 引用

configs/visdrone/ppyoloe_crn_l_80e_visdrone_largesize.yml → configs/visdrone/ppyoloe_crn_l_alpha_largesize_80e_visdrone.yml

 _BASE_: [
   'ppyoloe_crn_l_80e_visdrone.yml',
 ]
-weights: output/ppyoloe_crn_l_80e_visdrone_largesize/model_final
-
+weights: output/ppyoloe_crn_l_alpha_largesize_80e_visdrone/model_final
 pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams
 
-TrainReader:
-  batch_size: 2
+
+CSPResNet:
+  use_alpha: True
+
 
 LearningRate:
   base_lr: 0.0025

configs/visdrone/ppyoloe_p2_crn_l_80e_visdrone.yml → configs/visdrone/ppyoloe_crn_l_p2_alpha_80e_visdrone.yml

 _BASE_: [
   'ppyoloe_crn_l_80e_visdrone.yml',
 ]
-weights: output/ppyoloe_p2_crn_l_80e_visdrone/model_final
-
+weights: output/ppyoloe_crn_l_p2_alpha_80e_visdrone/model_final
 pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams
 
+
 TrainReader:
   batch_size: 4
 
 LearningRate:
   base_lr: 0.005
 
+
 CSPResNet:
   return_idx: [0, 1, 2, 3]
+  use_alpha: True
 
 CustomCSPPAN:
   out_channels: [768, 384, 192, 64]

configs/visdrone/ppyoloe_crn_l_p2_alpha_largesize_80e_visdrone.yml

+_BASE_: [
+  'ppyoloe_crn_l_80e_visdrone.yml',
+]
+weights: output/ppyoloe_crn_l_p2_alpha_largesize_80e_visdrone/model_final
+pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams
+
+
+LearningRate:
+  base_lr: 0.005
+
+
+CSPResNet:
+  return_idx: [0, 1, 2, 3]
+  use_alpha: True
+
+CustomCSPPAN:
+  out_channels: [768, 384, 192, 64]
+
+PPYOLOEHead:
+  fpn_strides: [32, 16, 8, 4]
+
+
+worker_num: 2
+eval_height: &eval_height 1920
+eval_width: &eval_width 1920
+eval_size: &eval_size [*eval_height, *eval_width]
+
+TrainReader:
+  sample_transforms:
+    - Decode: {}
+    - RandomDistort: {}
+    - RandomExpand: {fill_value: [123.675, 116.28, 103.53]}
+    - RandomCrop: {}
+    - RandomFlip: {}
+  batch_transforms:
+    - BatchRandomResize: {target_size: [1024, 1088, 1152, 1216, 1280, 1344, 1408, 1472, 1536, 1600, 1664, 1728, 1792, 1856, 1920, 1984, 2048], random_size: True, random_interp: True, keep_ratio: False}
+    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
+    - Permute: {}
+    - PadGT: {}
+  batch_size: 1
+  shuffle: true
+  drop_last: true
+  use_shared_memory: true
+  collate_batch: true
+
+EvalReader:
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: *eval_size, keep_ratio: False, interp: 2}
+    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
+    - Permute: {}
+  batch_size: 2
+
+TestReader:
+  inputs_def:
+    image_shape: [3, *eval_height, *eval_width]
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: *eval_size, keep_ratio: False, interp: 2}
+    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
+    - Permute: {}
+  batch_size: 1

configs/visdrone/ppyoloe_crn_s_p2_alpha_80e_visdrone.yml

+_BASE_: [
+  'ppyoloe_crn_s_80e_visdrone.yml',
+]
+weights: output/ppyoloe_crn_s_p2_alpha_80e_visdrone/model_final
+
+pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams
+
+TrainReader:
+  batch_size: 4
+
+LearningRate:
+  base_lr: 0.005
+
+CSPResNet:
+  return_idx: [0, 1, 2, 3]
+  use_alpha: True
+
+CustomCSPPAN:
+  out_channels: [768, 384, 192, 64]
+
+PPYOLOEHead:
+  fpn_strides: [32, 16, 8, 4]