update res2net_v1b

Author: gasvn

README.md

@@ -1,5 +1,7 @@
 # Res2Net for Pose Estimation
 
+## Update
+- [2020.3.13] Res2Net_v1b based Pose Estimation results are released now.
 
 ## Introduction
 This repo uses [*Simple Baselines*](http://openaccess.thecvf.com/content_ECCV_2018/html/Bin_Xiao_Simple_Baselines_for_ECCV_2018_paper.html) as the baseline method for Pose Estimation. 
@@ -13,12 +15,16 @@ More detail can be found on [ "Res2Net: A New Multi-scale Backbone Architecture"
 | Arch                      |Person detector | Input size   |   AP  | Ap .5 | AP .75 | AP (M) | AP (L) |
 |---------------------------|----------------|--------------|-------|-------|--------|--------|--------|
 | pose_resnet_50            | prdbox         |    256x192   | 0.704 | 0.886 |  0.783 |  0.671 |  0.772 |
-| **pose_res2net_50**       | prdbox         |    256x192   | 0.715 | 0.890 |  0.793 |  0.682 |  0.784 |
+| pose_res2net_50           | prdbox         |    256x192   | 0.715 | 0.890 |  0.793 |  0.682 |  0.784 |
 | pose_resnet_50            | GTbox          |    256x192   | 0.724 | 0.915 |  0.804 |  0.697 |  0.765 | 
-| **pose_res2net_50**       | GTbox          |    256x192   | 0.737 | 0.925 |  0.814 |  0.708 |  0.782 |
+| pose_res2net_50           | GTbox          |    256x192   | 0.737 | 0.925 |  0.814 |  0.708 |  0.782 |
 | pose_resnet_101           | prdbox         |    256x192   | 0.714 | 0.893 |  0.793 |  0.681 |  0.781 |
-| **pose_res2net_101**      | prdbox         |    256x192   | 0.722 | 0.894 |  0.798 |  0.689 |  0.792 |
-| **pose_res2net_101**      | GTbox          |    256x192   | 0.744 | 0.926 |  0.826 |  0.720 |  0.785 |
+| pose_res2net_101          | prdbox         |    256x192   | 0.722 | 0.894 |  0.798 |  0.689 |  0.792 |
+| pose_res2net_101          | GTbox          |    256x192   | 0.744 | 0.926 |  0.826 |  0.720 |  0.785 |
+| **pose_res2net_v1b_50**   | prdbox         |    256x192   | 0.722 | 0.895 |  0.797 |  0.685 |  0.794 |
+| **pose_res2net_v1b_50**   | GTbox          |    256x192   | 0.743 | 0.926 |  0.816 |  0.713 |  0.792 |
+| **pose_res2net_101**      | prdbox         |    256x192   | 0.730 | 0.895 |  0.803 |  0.695 |  0.800 |
+| **pose_res2net_101**      | GTbox          |    256x192   | 0.753 | 0.926 |  0.825 |  0.722 |  0.801 |
 
 
 ### Note:

experiments/coco/res2net/res2net101_v1b_4s_26w_256x192_d256x3_adam_lr1e-3.yaml

@@ -0,0 +1,85 @@
+AUTO_RESUME: true
+CUDNN:
+  BENCHMARK: true
+  DETERMINISTIC: false
+  ENABLED: true
+DATA_DIR: ''
+GPUS: (0,1,2,3)
+OUTPUT_DIR: 'output'
+LOG_DIR: 'log'
+WORKERS: 24
+PRINT_FREQ: 100
+
+DATASET:
+  COLOR_RGB: false
+  DATASET: 'coco'
+  ROOT: 'data/coco/'
+  TEST_SET: 'val2017'
+  TRAIN_SET: 'train2017'
+  FLIP: true
+  ROT_FACTOR: 40
+  SCALE_FACTOR: 0.3
+MODEL:
+  NAME: 'pose_res2net_v1b'
+  SCALE: 4
+  BASEWIDTH: 26
+  PRETRAINED: '/home/shgao/.torch/models/res2net101_v1b_26w_4s-0812c246.pth'
+  IMAGE_SIZE:
+  - 192
+  - 256
+  HEATMAP_SIZE:
+  - 48
+  - 64
+  SIGMA: 2
+  NUM_JOINTS: 17
+  TARGET_TYPE: 'gaussian'
+  EXTRA:
+    FINAL_CONV_KERNEL: 1
+    DECONV_WITH_BIAS: false
+    NUM_DECONV_LAYERS: 3
+    NUM_DECONV_FILTERS:
+    - 256
+    - 256
+    - 256
+    NUM_DECONV_KERNELS:
+    - 4
+    - 4
+    - 4
+    NUM_LAYERS: 101
+LOSS:
+  USE_TARGET_WEIGHT: true
+TRAIN:
+  BATCH_SIZE_PER_GPU: 32
+  SHUFFLE: true
+  BEGIN_EPOCH: 0
+  END_EPOCH: 140
+  OPTIMIZER: 'adam'
+  LR: 0.001
+  LR_FACTOR: 0.1
+  LR_STEP:
+  - 90
+  - 120
+  WD: 0.0001
+  GAMMA1: 0.99
+  GAMMA2: 0.0
+  MOMENTUM: 0.9
+  NESTEROV: false
+TEST:
+  BATCH_SIZE_PER_GPU: 32
+  COCO_BBOX_FILE: 'data/coco/person_detection_results/COCO_val2017_detections_AP_H_56_person.json'
+  BBOX_THRE: 1.0
+  IMAGE_THRE: 0.0
+  IN_VIS_THRE: 0.2
+  MODEL_FILE: ''
+  NMS_THRE: 1.0
+  OKS_THRE: 0.9
+  FLIP_TEST: true
+  POST_PROCESS: true
+  SHIFT_HEATMAP: true
+  USE_GT_BBOX: true
+DEBUG:
+  DEBUG: true
+  SAVE_BATCH_IMAGES_GT: true
+  SAVE_BATCH_IMAGES_PRED: true
+  SAVE_HEATMAPS_GT: true
+  SAVE_HEATMAPS_PRED: true

experiments/coco/res2net/res2net50_v1b_4s_26w_256x192_d256x3_adam_lr1e-3.yaml

@@ -0,0 +1,85 @@
+AUTO_RESUME: true
+CUDNN:
+  BENCHMARK: true
+  DETERMINISTIC: false
+  ENABLED: true
+DATA_DIR: ''
+GPUS: (0,1,2,3)
+OUTPUT_DIR: 'output'
+LOG_DIR: 'log'
+WORKERS: 24
+PRINT_FREQ: 100
+
+DATASET:
+  COLOR_RGB: false
+  DATASET: 'coco'
+  ROOT: 'data/coco/'
+  TEST_SET: 'val2017'
+  TRAIN_SET: 'train2017'
+  FLIP: true
+  ROT_FACTOR: 40
+  SCALE_FACTOR: 0.3
+MODEL:
+  NAME: 'pose_res2net_v1b'
+  SCALE: 4
+  BASEWIDTH: 26
+  PRETRAINED: '/home/shgao/.torch/models/res2net50_v1b_26w_4s-3cf99910.pth'
+  IMAGE_SIZE:
+  - 192
+  - 256
+  HEATMAP_SIZE:
+  - 48
+  - 64
+  SIGMA: 2
+  NUM_JOINTS: 17
+  TARGET_TYPE: 'gaussian'
+  EXTRA:
+    FINAL_CONV_KERNEL: 1
+    DECONV_WITH_BIAS: false
+    NUM_DECONV_LAYERS: 3
+    NUM_DECONV_FILTERS:
+    - 256
+    - 256
+    - 256
+    NUM_DECONV_KERNELS:
+    - 4
+    - 4
+    - 4
+    NUM_LAYERS: 50
+LOSS:
+  USE_TARGET_WEIGHT: true
+TRAIN:
+  BATCH_SIZE_PER_GPU: 32
+  SHUFFLE: true
+  BEGIN_EPOCH: 0
+  END_EPOCH: 140
+  OPTIMIZER: 'adam'
+  LR: 0.001
+  LR_FACTOR: 0.1
+  LR_STEP:
+  - 90
+  - 120
+  WD: 0.0001
+  GAMMA1: 0.99
+  GAMMA2: 0.0
+  MOMENTUM: 0.9
+  NESTEROV: false
+TEST:
+  BATCH_SIZE_PER_GPU: 32
+  COCO_BBOX_FILE: 'data/coco/person_detection_results/COCO_val2017_detections_AP_H_56_person.json'
+  BBOX_THRE: 1.0
+  IMAGE_THRE: 0.0
+  IN_VIS_THRE: 0.2
+  MODEL_FILE: ''
+  NMS_THRE: 1.0
+  OKS_THRE: 0.9
+  FLIP_TEST: true
+  POST_PROCESS: true
+  SHIFT_HEATMAP: true
+  USE_GT_BBOX: true
+DEBUG:
+  DEBUG: true
+  SAVE_BATCH_IMAGES_GT: true
+  SAVE_BATCH_IMAGES_PRED: true
+  SAVE_HEATMAPS_GT: true
+  SAVE_HEATMAPS_PRED: true

lib/models/__init__.py

@@ -14,3 +14,4 @@
 
 import models.pose_resnet
 import models.pose_res2net
+import models.pose_res2net_v1b