mmpose_RTDT-1550_Train_a_network_in_LS_based_on_merged_annotation

.drone.yml

@@ -0,0 +1,102 @@
+---
+#######################################################################################################################
+#######################################################################################################################
+#######################################################################################################################
+#######################         Jobs below are used to build only             #########################################
+#######################        master  branches for all containers            #########################################
+#######################                                                       #########################################
+#######################################################################################################################
+#######################################################################################################################
+#######################################################################################################################
+
+# This pipeline would be used after each merge request to build mmpose latest docker images
+kind: pipeline
+type: docker
+name: mmpose LS_mmpose_latest
+
+platform:
+  arch: amd64
+  os: linux
+
+trigger:
+  branch:
+    - main
+  event:
+    - push
+
+clone:
+  depth: 1
+
+steps:
+  - name: Build latest LS_mmpose docker image
+    image: plugins/docker:20.14
+    environment:
+      DOCKER_BUILDKIT: 1
+    settings: 
+      dockerfile: docker/LabelStudio.Dockerfile
+      context: docker/
+      registry: quay.io
+      repo: quay.io/logivations/ml_all
+      privileged: true
+      build_args:
+        - BUILDKIT_INLINE_CACHE=1
+      cache_from: quay.io/logivations/ml_all:LS_mmpose_latest
+      tags:
+        - LS_mmpose_latest
+        - LS_mmpose_latest_${DRONE_COMMIT_SHA}
+      username:
+        from_secret: DOCKER_QUAY_USERNAME
+      password:
+        from_secret: DOCKER_QUAY_PASSWORD
+
+#######################################################################################################################
+#######################################################################################################################
+#######################################################################################################################
+#######################              Jobs below are used to run               #########################################
+#######################             pull request validation only              #########################################
+#######################                                                       #########################################
+#######################################################################################################################
+#######################################################################################################################
+#######################################################################################################################
+
+# 2. Build  PR docker image for mmpose
+---
+kind: pipeline
+type: docker
+name: PR validation build mmpose images
+
+platform:
+  arch: amd64
+  os: linux
+
+trigger:
+  event:
+    include:
+      - pull_request
+
+clone: 
+  depth: 50
+
+steps:
+  - name: Build LS_mmpose docker image for pull request
+    image: plugins/docker:20.14
+    environment:
+      DOCKER_BUILDKIT: 1
+    settings: 
+      dockerfile: docker/LabelStudio.Dockerfile
+      context: docker/
+      registry: quay.io
+      repo: quay.io/logivations/ml_all
+      privileged: true
+      build_args:
+        - BUILDKIT_INLINE_CACHE=1
+      cache_from: 
+        - quay.io/logivations/ml_all:LS_mmpose_latest
+        - quay.io/logivations/ml_all:LS_mmpose_pr${DRONE_PULL_REQUEST}
+      tags:
+        - LS_mmpose_pr${DRONE_PULL_REQUEST}
+        - LS_mmpose_pr${DRONE_PULL_REQUEST}_${DRONE_COMMIT_SHA}
+      username:
+        from_secret: DOCKER_QUAY_USERNAME
+      password:
+        from_secret: DOCKER_QUAY_PASSWORD

.pre-commit-config.yaml

@@ -13,7 +13,7 @@ repos:
     hooks:
       - id: isort
   - repo: https://github.com/pre-commit/mirrors-yapf
-    rev: v0.30.0
+    rev: v0.40.1
     hooks:
       - id: yapf
   - repo: https://github.com/pre-commit/pre-commit-hooks

auto_training/config_factories/mmpose_config_factory.py

@@ -0,0 +1,170 @@
+from mmcv import Config
+
+
+def make_mmpose_cfg(
+        work_dir: str,
+        labels: list = []
+):
+    channel_cfg = dict(
+        num_output_channels=len(labels),
+        dataset_joints=len(labels),
+        dataset_channel=list(range(len(labels))),
+        inference_channel=list(range(len(labels)))
+    )
+    val_pipeline = [
+        dict(type='LoadImageFromFile'),
+        dict(type='TopDownAffine'),
+        dict(type='ToTensor'),
+        dict(
+            type='NormalizeTensor',
+            mean=[0.485, 0.456, 0.406],
+            std=[0.229, 0.224, 0.225]),
+        dict(
+            type='Collect',
+            keys=['img'],
+            meta_keys=[
+                'image_file', 'center', 'scale', 'rotation', 'bbox_score',
+                'flip_pairs'
+            ]),
+    ]
+    train_pipeline = [
+        dict(type='LoadImageFromFile'),
+        dict(type='TopDownRandomFlip', flip_prob=0.5),
+        dict(
+            type='TopDownGetRandomScaleRotation', rot_factor=180, scale_factor=0.5, rot_prob=0.9),
+        dict(type='TopDownAffine'),
+        dict(type='ToTensor'),
+        dict(
+            type='NormalizeTensor',
+            mean=[0.485, 0.456, 0.406],
+            std=[0.229, 0.224, 0.225]),
+        dict(type='TopDownGenerateTarget', sigma=3),
+        dict(
+            type='Collect',
+            keys=['img', 'target', 'target_weight'],
+            meta_keys=[
+                'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
+                'rotation', 'bbox_score', 'flip_pairs'
+            ]),
+    ]
+    data_cfg = dict(
+        image_size=[192, 256],
+        heatmap_size=[48, 64],
+        num_output_channels=channel_cfg['num_output_channels'],
+        num_joints=channel_cfg['dataset_joints'],
+        dataset_channel=channel_cfg['dataset_channel'],
+        inference_channel=channel_cfg['inference_channel'],
+        soft_nms=False,
+        nms_thr=1.0,
+        oks_thr=0.9,
+        vis_thr=0.2,
+        use_gt_bbox=True,
+        det_bbox_thr=0.0,
+        bbox_file=f'/dataset/train/coco_train.json',
+    )
+
+    cfg = dict(
+        work_dir=work_dir,
+        total_epochs=100,
+        default_scope='mmpose',
+        optimizer_config=dict(grad_clip=None),
+        log_level='INFO',
+        load_from=None,
+        resume_from=None,
+        dist_params=dict(backend='nccl'),
+        optimizer=dict(
+            type='Adam',
+            lr=1e-4,
+        ),
+        workflow=[('train', 1)],
+        checkpoint_config=dict(interval=100),
+        evaluation=dict(interval=10, metric='mAP', key_indicator='AP'),
+        lr_config=dict(
+            policy='step',
+            warmup='linear',
+            warmup_iters=5,
+            warmup_ratio=0.001,
+            step=[360, 380]
+        ),
+        log_config=dict(
+            interval=25,
+            hooks=[
+                dict(type='TextLoggerHook'),
+                dict(type='TensorboardLoggerHook')
+            ]
+        ),
+        default_hooks=dict(
+            # record the time of every iteration.
+            timer=dict(type='EpochTimerHook'),
+
+            # print log every 10 iterations.
+            logger=dict(type='LoggerHook', interval=10),
+
+            # enable the parameter scheduler.
+            param_scheduler=dict(type='ParamSchedulerHook'),
+
+            # save checkpoint per epoch.
+            checkpoint=dict(type='CheckpointHook', interval=100),
+
+            # set sampler seed in distributed evrionment.
+            sampler_seed=dict(type='DistSamplerSeedHook'),
+
+            # validation results visualization, set True to enable it.
+            visualization=dict(type='VisualizationHook', enable=True),
+        ),
+        val_pipeline=val_pipeline,
+        test_pipeline=val_pipeline,
+        data=dict(
+            samples_per_gpu=8,
+            workers_per_gpu=2,
+            train=dict(
+                type='LiftedForkDatasetAnyKP',
+                ann_file='/dataset/annotations/coco_train.json',
+                img_prefix='/dataset/train/',
+                data_cfg=data_cfg,
+                pipeline=train_pipeline,
+                num_joints=len(labels)
+            ),
+            val=dict(
+                type='LiftedForkDatasetAnyKP',
+                ann_file='/dataset/annotations/coco_val.json',
+                img_prefix='/dataset/val/',
+                data_cfg=data_cfg,
+                pipeline=val_pipeline,
+                num_joints=len(labels)
+            ),
+            test=dict(
+                type='LiftedForkDatasetAnyKP',
+                ann_file='/dataset/annotations/coco_test.json',
+                img_prefix='/dataset/test/',
+                data_cfg=data_cfg,
+                pipeline=val_pipeline,
+                num_joints=len(labels)
+            ),
+        ),
+        val_evaluator=[
+            dict(prefix='val', topk=(1,), type='Accuracy'),
+            dict(prefix='val', type='AveragePrecision'),
+            dict(prefix='val', type='SingleLabelMetric'),
+        ],
+        model=dict(
+            type='TopDown',
+            pretrained='torchvision://resnet18',
+            backbone=dict(type='ResNet', depth=18),
+            keypoint_head=dict(
+                type='TopDownSimpleHead',
+                in_channels=512,
+                out_channels=channel_cfg['num_output_channels'],
+                loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
+            train_cfg=dict(by_epoch=True),
+            test_cfg=dict(
+                flip_test=True,
+                post_process='default',
+                shift_heatmap=True,
+                modulate_kernel=11
+            )
+        )
+
+    )
+    cfg = Config(cfg)
+    return cfg

auto_training/inference.py

@@ -0,0 +1,93 @@
+from pathlib import Path
+from argparse import ArgumentParser
+import os
+import json
+import time
+from mmpose.apis.custom_inferencers import MMPoseInferencer
+
+class FPSLogger:
+    def __init__(self, num_of_images):
+        self.tottime = 0.0
+        self.count = 0
+        self.last_record = 0.0
+        self.last_print = time.time()
+        self.interval = 3
+        self.num_of_images = num_of_images
+
+    def start_record(self):
+        self.last_record = time.time()
+
+    def end_record(self):
+        self.tottime += time.time() - self.last_record
+        self.count += 1
+        self.print_fps()
+
+    def print_fps(self):
+        if time.time() - self.last_print > self.interval:
+            print(f"{time.strftime('%Y-%m-%d %H:%M:%S')} - mmpret - INFO - Predict({self.count}/{self.num_of_images}) "
+                  f"- Inference running at {self.count / self.tottime:.3f} FPS")
+            self.last_print = time.time()
+
+def main(args):
+    fps_logger = FPSLogger(len(os.listdir(args.images_dir)))
+    inference = MMPoseInferencer(
+        config=args.config,
+        pretrained=args.checkpoint,
+        bboxes_filepath=args.bboxes_filepath,
+    )
+    print(f"Inference classes: {args.classes}")
+    # if args.silent:
+    #     inference.show_progress = False
+
+    images: Path = args.images_dir
+    output_dir: Path = args.output_dir
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    start_time = time.time()
+    image_paths = list(images.glob("**/*.jpg"))
+    for i, p in enumerate(image_paths, 1):
+        try:
+            fps_logger.start_record()
+            result = inference(str(p))
+            fps_logger.end_record()
+
+            file_name = str(p).split("/")[-1].replace(".jpg", ".json")
+            pred_path = os.path.join(output_dir, file_name)
+            prediction = {
+                "result": [
+                    {
+                        "type": "choices",
+                        "value": {"choices": [result[0]['pred_class']]},
+                        "origin": "manual",
+                        "to_name": "image",
+                        "from_name": "choice",
+                    }
+                ],
+            }
+
+            with open(pred_path, "w") as f:
+                json.dump(prediction, f)
+        except Exception as e:
+            print(f"Failed with {p}. {e}")
+    print(f"Inference time: {round(time.time() - start_time, 2)} s.")
+
+
+if __name__ == "__main__":
+    parser = ArgumentParser()
+    parser.add_argument("checkpoint")
+    parser.add_argument("config")
+    parser.add_argument("images_dir", type=Path)
+    parser.add_argument("output_dir", type=Path)
+    parser.add_argument("bboxes_filepath", type=Path)
+    parser.add_argument(
+        "--silent",
+        action="store_true",
+        help="suppress progress bars and verbose output")
+    parser.add_argument(
+        '--classes',
+        nargs='+',
+        required=True,
+        help='list of classes for the training'
+    )
+    config = parser.parse_args()
+    main(config)