更换文档检测模型

2024-08-27 14:42:45 +08:00
parent aea6f19951
commit 1514e09c40
2072 changed files with 254336 additions and 4967 deletions
--- a/paddle_detection/configs/keypoint/higherhrnet/higherhrnet_hrnet_w32_512.yml
+++ b/paddle_detection/configs/keypoint/higherhrnet/higherhrnet_hrnet_w32_512.yml
@@ -0,0 +1,139 @@
+use_gpu: true
+log_iter: 10
+save_dir: output
+snapshot_epoch: 10
+weights: output/higherhrnet_hrnet_w32_512/model_final
+epoch: 300
+num_joints: &num_joints 17
+flip_perm: &flip_perm [0, 2, 1, 4, 3, 6, 5, 8, 7, 10, 9, 12, 11, 14, 13, 16, 15]
+input_size: &input_size 512
+hm_size: &hm_size 128
+hm_size_2x: &hm_size_2x 256
+max_people: &max_people 30
+metric: COCO
+IouType: keypoints
+num_classes: 1
+
+
+#####model
+architecture: HigherHRNet
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/Trunc_HRNet_W32_C_pretrained.pdparams
+
+HigherHRNet:
+  backbone: HRNet
+  hrhrnet_head: HrHRNetHead
+  post_process: HrHRNetPostProcess
+  flip_perm: *flip_perm
+  eval_flip: true
+
+HRNet:
+  width: &width 32
+  freeze_at: -1
+  freeze_norm: false
+  return_idx: [0]
+
+HrHRNetHead:
+  num_joints: *num_joints
+  width: *width
+  loss: HrHRNetLoss
+  swahr: false
+
+HrHRNetLoss:
+  num_joints: *num_joints
+  swahr: false
+
+
+#####optimizer
+LearningRate:
+  base_lr: 0.001
+  schedulers:
+  - !PiecewiseDecay
+    milestones: [200, 260]
+    gamma: 0.1
+  - !LinearWarmup
+    start_factor: 0.001
+    steps: 1000
+
+OptimizerBuilder:
+  optimizer:
+    type: Adam
+  regularizer: None
+
+#####data
+TrainDataset:
+  !KeypointBottomUpCocoDataset
+    image_dir: train2017
+    anno_path: annotations/person_keypoints_train2017.json
+    dataset_dir: dataset/coco
+    num_joints: *num_joints
+    return_bbox: False
+    return_area: False
+    return_class: False
+
+EvalDataset:
+  !KeypointBottomUpCocoDataset
+    image_dir: val2017
+    anno_path: annotations/person_keypoints_val2017.json
+    dataset_dir: dataset/coco
+    num_joints: *num_joints
+    test_mode: true
+    return_bbox: False
+    return_area: False
+    return_class: False
+
+TestDataset:
+  !ImageFolder
+    anno_path: dataset/coco/keypoint_imagelist.txt
+
+worker_num: 8
+global_mean: &global_mean [0.485, 0.456, 0.406]
+global_std: &global_std [0.229, 0.224, 0.225]
+TrainReader:
+  sample_transforms:
+    - RandomAffine:
+        max_degree: 30
+        scale: [0.75, 1.5]
+        max_shift: 0.2
+        trainsize: [*input_size, *input_size]
+        hmsize: [*hm_size, *hm_size_2x]
+    - KeyPointFlip:
+        flip_prob: 0.5
+        flip_permutation: *flip_perm
+        hmsize: [*hm_size, *hm_size_2x]
+    - ToHeatmaps:
+        num_joints: *num_joints
+        hmsize: [*hm_size, *hm_size_2x]
+        sigma: 2
+    - TagGenerate:
+        num_joints: *num_joints
+        max_people: *max_people
+    - NormalizePermute:
+        mean: *global_mean
+        std: *global_std
+  batch_size: 20
+  shuffle: true
+  drop_last: true
+  use_shared_memory: true
+
+EvalReader:
+  sample_transforms:
+    - EvalAffine:
+        size: *input_size
+    - NormalizeImage:
+        mean: *global_mean
+        std: *global_std
+        is_scale: true
+    - Permute: {}
+  batch_size: 1
+
+TestReader:
+  sample_transforms:
+    - Decode: {}
+    - EvalAffine:
+        size: *input_size
+    - NormalizeImage:
+        mean: *global_mean
+        std: *global_std
+        is_scale: true
+    - Permute: {}
+  batch_size: 1
--- a/paddle_detection/configs/keypoint/higherhrnet/higherhrnet_hrnet_w32_512_swahr.yml
+++ b/paddle_detection/configs/keypoint/higherhrnet/higherhrnet_hrnet_w32_512_swahr.yml
@@ -0,0 +1,140 @@
+use_gpu: true
+log_iter: 10
+save_dir: output
+snapshot_epoch: 10
+weights: output/higherhrnet_hrnet_w32_512_swahr/model_final
+epoch: 300
+num_joints: &num_joints 17
+flip_perm: &flip_perm [0, 2, 1, 4, 3, 6, 5, 8, 7, 10, 9, 12, 11, 14, 13, 16, 15]
+input_size: &input_size 512
+hm_size: &hm_size 128
+hm_size_2x: &hm_size_2x 256
+max_people: &max_people 30
+metric: COCO
+IouType: keypoints
+num_classes: 1
+
+
+#####model
+architecture: HigherHRNet
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/Trunc_HRNet_W32_C_pretrained.pdparams
+
+HigherHRNet:
+  backbone: HRNet
+  hrhrnet_head: HrHRNetHead
+  post_process: HrHRNetPostProcess
+  flip_perm: *flip_perm
+  eval_flip: true
+
+HRNet:
+  width: &width 32
+  freeze_at: -1
+  freeze_norm: false
+  return_idx: [0]
+
+HrHRNetHead:
+  num_joints: *num_joints
+  width: *width
+  loss: HrHRNetLoss
+  swahr: true
+
+HrHRNetLoss:
+  num_joints: *num_joints
+  swahr: true
+
+
+#####optimizer
+LearningRate:
+  base_lr: 0.001
+  schedulers:
+  - !PiecewiseDecay
+    milestones: [200, 260]
+    gamma: 0.1
+  - !LinearWarmup
+    start_factor: 0.001
+    steps: 1000
+
+OptimizerBuilder:
+  optimizer:
+    type: Adam
+  regularizer: None
+
+
+#####data
+TrainDataset:
+  !KeypointBottomUpCocoDataset
+    image_dir: train2017
+    anno_path: annotations/person_keypoints_train2017.json
+    dataset_dir: dataset/coco
+    num_joints: *num_joints
+    return_bbox: False
+    return_area: False
+    return_class: False
+
+EvalDataset:
+  !KeypointBottomUpCocoDataset
+    image_dir: val2017
+    anno_path: annotations/person_keypoints_val2017.json
+    dataset_dir: dataset/coco
+    num_joints: *num_joints
+    test_mode: true
+    return_bbox: False
+    return_area: False
+    return_class: False
+
+TestDataset:
+  !ImageFolder
+    anno_path: dataset/coco/keypoint_imagelist.txt
+
+worker_num: 8
+global_mean: &global_mean [0.485, 0.456, 0.406]
+global_std: &global_std [0.229, 0.224, 0.225]
+TrainReader:
+  sample_transforms:
+    - RandomAffine:
+        max_degree: 30
+        scale: [0.75, 1.5]
+        max_shift: 0.2
+        trainsize: [*input_size, *input_size]
+        hmsize: [*hm_size, *hm_size_2x]
+    - KeyPointFlip:
+        flip_prob: 0.5
+        flip_permutation: *flip_perm
+        hmsize: [*hm_size, *hm_size_2x]
+    - ToHeatmaps:
+        num_joints: *num_joints
+        hmsize: [*hm_size, *hm_size_2x]
+        sigma: 2
+    - TagGenerate:
+        num_joints: *num_joints
+        max_people: *max_people
+    - NormalizePermute:
+        mean: *global_mean
+        std: *global_std
+  batch_size: 16
+  shuffle: true
+  drop_last: true
+  use_shared_memory: true
+
+EvalReader:
+  sample_transforms:
+    - EvalAffine:
+        size: *input_size
+    - NormalizeImage:
+        mean: *global_mean
+        std: *global_std
+        is_scale: true
+    - Permute: {}
+  batch_size: 1
+
+TestReader:
+  sample_transforms:
+    - Decode: {}
+    - EvalAffine:
+        size: *input_size
+    - NormalizeImage:
+        mean: *global_mean
+        std: *global_std
+        is_scale: true
+    - Permute: {}
+  batch_size: 1
--- a/paddle_detection/configs/keypoint/higherhrnet/higherhrnet_hrnet_w32_640.yml
+++ b/paddle_detection/configs/keypoint/higherhrnet/higherhrnet_hrnet_w32_640.yml
@@ -0,0 +1,139 @@
+use_gpu: true
+log_iter: 10
+save_dir: output
+snapshot_epoch: 10
+weights: output/higherhrnet_hrnet_w32_640/model_final
+epoch: 300
+num_joints: &num_joints 17
+flip_perm: &flip_perm [0, 2, 1, 4, 3, 6, 5, 8, 7, 10, 9, 12, 11, 14, 13, 16, 15]
+input_size: &input_size 640
+hm_size: &hm_size 160
+hm_size_2x: &hm_size_2x 320
+max_people: &max_people 30
+metric: COCO
+IouType: keypoints
+num_classes: 1
+
+
+#####model
+architecture: HigherHRNet
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/Trunc_HRNet_W32_C_pretrained.pdparams
+
+HigherHRNet:
+  backbone: HRNet
+  hrhrnet_head: HrHRNetHead
+  post_process: HrHRNetPostProcess
+  flip_perm: *flip_perm
+  eval_flip: true
+
+HRNet:
+  width: &width 32
+  freeze_at: -1
+  freeze_norm: false
+  return_idx: [0]
+
+HrHRNetHead:
+  num_joints: *num_joints
+  width: *width
+  loss: HrHRNetLoss
+  swahr: false
+
+HrHRNetLoss:
+  num_joints: *num_joints
+  swahr: false
+
+
+#####optimizer
+LearningRate:
+  base_lr: 0.001
+  schedulers:
+  - !PiecewiseDecay
+    milestones: [200, 260]
+    gamma: 0.1
+  - !LinearWarmup
+    start_factor: 0.001
+    steps: 1000
+
+OptimizerBuilder:
+  optimizer:
+    type: Adam
+  regularizer: None
+
+#####data
+TrainDataset:
+  !KeypointBottomUpCocoDataset
+    image_dir: train2017
+    anno_path: annotations/person_keypoints_train2017.json
+    dataset_dir: dataset/coco
+    num_joints: *num_joints
+    return_bbox: False
+    return_area: False
+    return_class: False
+
+EvalDataset:
+  !KeypointBottomUpCocoDataset
+    image_dir: val2017
+    anno_path: annotations/person_keypoints_val2017.json
+    dataset_dir: dataset/coco
+    num_joints: *num_joints
+    test_mode: true
+    return_bbox: False
+    return_area: False
+    return_class: False
+
+TestDataset:
+  !ImageFolder
+    anno_path: dataset/coco/keypoint_imagelist.txt
+
+worker_num: 8
+global_mean: &global_mean [0.485, 0.456, 0.406]
+global_std: &global_std [0.229, 0.224, 0.225]
+TrainReader:
+  sample_transforms:
+    - RandomAffine:
+        max_degree: 30
+        scale: [0.75, 1.5]
+        max_shift: 0.2
+        trainsize: [*input_size, *input_size]
+        hmsize: [*hm_size, *hm_size_2x]
+    - KeyPointFlip:
+        flip_prob: 0.5
+        flip_permutation: *flip_perm
+        hmsize: [*hm_size, *hm_size_2x]
+    - ToHeatmaps:
+        num_joints: *num_joints
+        hmsize: [*hm_size, *hm_size_2x]
+        sigma: 2
+    - TagGenerate:
+        num_joints: *num_joints
+        max_people: *max_people
+    - NormalizePermute:
+        mean: *global_mean
+        std: *global_std
+  batch_size: 20
+  shuffle: true
+  drop_last: true
+  use_shared_memory: true
+
+EvalReader:
+  sample_transforms:
+    - EvalAffine:
+        size: *input_size
+    - NormalizeImage:
+        mean: *global_mean
+        std: *global_std
+        is_scale: true
+    - Permute: {}
+  batch_size: 1
+
+TestReader:
+  sample_transforms:
+    - Decode: {}
+    - EvalAffine:
+        size: *input_size
+    - NormalizeImage:
+        mean: *global_mean
+        std: *global_std
+        is_scale: true
+    - Permute: {}
+  batch_size: 1