Feedback

DDQ_seungsoo/README.md

@@ -0,0 +1,56 @@
+## Copy Paste 활용
+1. `custom`폴더의 `copy_paste_aug` 폴더를 `mmdetection/mmdet`에 넣어줍니다.
+2. `mmdet/datasets/transforms/__init__.py`에 아래와 같은 두 코드를 넣어줍니다.
+   ```
+   from mmdet.copy_paste_aug.custom_loader import LoadAnnotationswithbox
+   from mmdet.copy_paste_aug.custom_copypaste import custom_CopyPaste
+   ```
+3. `__all__` 리스트에 `'custom_CopyPaste','LoadAnnotationswithbox'` 추가
+
+## Train
+1. `train.py`의 train_pipeline 수정
+   ```
+   train_pipeline = [
+    dict(
+        type='custom_mosaic_copy_paste',paste_by_box=True),
+    dict(
+        type='Albu',
+        transforms=albu_train_transforms,
+        bbox_params=dict(
+            type='BboxParams',
+            format='pascal_voc',
+            label_fields=['gt_bboxes_labels', 'gt_ignore_flags'],
+            min_visibility=0.0,
+            filter_lost_elements=True),
+        keymap={
+            'img': 'image',
+            'gt_bboxes': 'bboxes'
+        },
+        skip_img_without_anno=True),
+    dict(type='FilterAnnotations', min_gt_bbox_wh=(50, 50), keep_empty=True),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type = 'RandomCrop',crop_size=img_scale),
+    dict(type='PackDetInputs')
+    ]
+   ```
+2. `train.py`의 train_dataset 수정
+   ```
+   train_dataset = dict(
+        _delete_=True,
+    # use MultiImageMixDataset wrapper to support mosaic and mixup
+    type='MultiImageMixDataset',
+    dataset=dict(
+        type='CocoDataset',
+        metainfo = metainfo,
+        data_root=data_root,
+        ann_file='train_2_annot.json',
+        data_prefix=dict(img= ''),
+        filter_cfg=dict(filter_empty_gt=True),
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', with_bbox=True)
+        ]),
+    pipeline=train_pipeline)
+   ```
+
+자세한 사항은 `ddq_train.py` 참고