YOLOv3 - Continue on #1695 (#3083)

* Implement YOLOv3 * Remove unused function * Update yolov3_ms_aug_273e.py Clean the comments in config file * Add README.md * port to mmdet-2.0 api * unify registry * port to ConvModule and remove ConvLayer * Refactor Backbone * Update README * Lint and format * Unify the class name * fix the `label - 1` problem * Move a lot hard-coded params to the __init__ function * Refactor YOLOV3Neck * Add norm_cfg and act_cfg to backbone * Update Config * Fix doc string * Fix nms (thanks to @LMerCy) * Add doc string * Update config * Remove pretrained in head and neck * Add support for conv_cfg in neck * Update mmdet/models/dense_heads/yolo_head.py Co-authored-by: Jerry Jiarui XU <xvjiarui0826@gmail.com> * Update mmdet/models/dense_heads/yolo_head.py Co-authored-by: Jerry Jiarui XU <xvjiarui0826@gmail.com> * Fix README.md * Fix typos * update config * flake8, yapf, docformatter, etc * Update README * Add conv_cfg to backbone and head * Move some config to arch_settings in backbone * Add doc strings and replace Warning with warnings.warn() * Fix bug. * Update doc * Add _frozen_stages for backbone * Update mmdet/models/backbones/darknet.py Co-authored-by: Jerry Jiarui XU <xvjiarui0826@gmail.com> * Fix inplace bug * fix indent * refactor config * set 8GPU lr * fixed typo * update performance table * Resolve conversation * Add anchor generator and coder * fixed test * Finish refactor * refactor anchor order * fixed batch size * Fixed train_cfg * fix yolo assigner * clean up * Fixed format * Update model zoo * change to mmcv pretrain link * add test forward * fixed comma and docstring * Refactor loss * reformat * fixed avg_factor * revert to original * fixed format * update table * fixed BCE Co-authored-by: Haoyu Wu <haoyu.wu@wdc.com> Co-authored-by: Haoyu Wu <wuhy08@users.noreply.github.com> Co-authored-by: Haoyu Wu <wuhaoyu1989@gmail.com> Co-authored-by: Jerry Jiarui XU <xvjiarui0826@gmail.com> Co-authored-by: xmpeng <1051323399@qq.com>
open-mmlab · Aug 31, 2020 · dfbb6d6 · dfbb6d6
1 parent 83f0ca4
commit dfbb6d6
Show file tree

Hide file tree

Showing 25 changed files with 1,540 additions and 13 deletions.
diff --git a/README.md b/README.md
@@ -97,6 +97,7 @@ Supported methods:
 - [x] [DetectoRS](configs/detectors/README.md)
 - [x] [Generalized Focal Loss](configs/gfl/README.md)
 - [x] [CornerNet](configs/cornernet/README.md)
+- [x] [YOLOv3](configs/yolo/README.md)
 
 Some other methods are also supported in [projects using MMDetection](./docs/projects.md).
 

diff --git a/configs/yolo/README.md b/configs/yolo/README.md
@@ -0,0 +1,25 @@
+# YOLOv3
+
+## Introduction
+```
+@misc{redmon2018yolov3,
+    title={YOLOv3: An Incremental Improvement},
+    author={Joseph Redmon and Ali Farhadi},
+    year={2018},
+    eprint={1804.02767},
+    archivePrefix={arXiv},
+    primaryClass={cs.CV}
+}
+```
+
+## Results and Models
+
+|    Backbone     |  Scale  | Lr schd | Mem (GB) | Inf time (fps) | box AP | Download  |
+| :-------------: | :-----: | :-----: | :------: | :------------: | :----: | :-------: |
+|   DarkNet-53    |   320   |   273e  |   2.7    |      63.9      |  27.9  | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmdetection/v2.0/yolo/yolov3_d53_320_273e_coco/yolov3_d53_320_273e_coco-421362b6.pth) &#124; [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmdetection/v2.0/yolo/yolov3_d53_320_273e_coco/yolov3_d53_320_273e_coco-20200819_172101.log.json) |
+|   DarkNet-53    |   416   |   273e  |   3.8    |      61.2      |  30.9  | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmdetection/v2.0/yolo/yolov3_d53_mstrain-416_273e_coco/yolov3_d53_mstrain-416_273e_coco-2b60fcd9.pth) &#124; [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmdetection/v2.0/yolo/yolov3_d53_mstrain-416_273e_coco/yolov3_d53_mstrain-416_273e_coco-20200819_173424.log.json) |
+|   DarkNet-53    |   608   |   273e  |   7.1    |      48.1      |  33.4  | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmdetection/v2.0/yolo/yolov3_d53_mstrain-608_273e_coco/yolov3_d53_mstrain-608_273e_coco-139f5633.pth) &#124; [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmdetection/v2.0/yolo/yolov3_d53_mstrain-608_273e_coco/yolov3_d53_mstrain-608_273e_coco-20200819_170820.log.json) |
+
+
+## Credit
+This implementation originates from the project of Haoyu Wu(@wuhy08) at Western Digital.
diff --git a/configs/yolo/yolov3_d53_320_273e_coco.py b/configs/yolo/yolov3_d53_320_273e_coco.py
@@ -0,0 +1,42 @@
+_base_ = './yolov3_d53_mstrain-608_273e_coco.py'
+# dataset settings
+img_norm_cfg = dict(mean=[0, 0, 0], std=[255., 255., 255.], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile', to_float32=True),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Expand',
+        mean=img_norm_cfg['mean'],
+        to_rgb=img_norm_cfg['to_rgb'],
+        ratio_range=(1, 2)),
+    dict(
+        type='MinIoURandomCrop',
+        min_ious=(0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
+        min_crop_size=0.3),
+    dict(type='Resize', img_scale=(320, 320), keep_ratio=True),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(320, 320),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='Pad', size_divisor=32),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    train=dict(pipeline=train_pipeline),
+    val=dict(pipeline=test_pipeline),
+    test=dict(pipeline=test_pipeline))
diff --git a/configs/yolo/yolov3_d53_mstrain-416_273e_coco.py b/configs/yolo/yolov3_d53_mstrain-416_273e_coco.py
@@ -0,0 +1,42 @@
+_base_ = './yolov3_d53_mstrain-608_273e_coco.py'
+# dataset settings
+img_norm_cfg = dict(mean=[0, 0, 0], std=[255., 255., 255.], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile', to_float32=True),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Expand',
+        mean=img_norm_cfg['mean'],
+        to_rgb=img_norm_cfg['to_rgb'],
+        ratio_range=(1, 2)),
+    dict(
+        type='MinIoURandomCrop',
+        min_ious=(0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
+        min_crop_size=0.3),
+    dict(type='Resize', img_scale=[(320, 320), (416, 416)], keep_ratio=True),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(416, 416),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='Pad', size_divisor=32),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    train=dict(pipeline=train_pipeline),
+    val=dict(pipeline=test_pipeline),
+    test=dict(pipeline=test_pipeline))
diff --git a/configs/yolo/yolov3_d53_mstrain-608_273e_coco.py b/configs/yolo/yolov3_d53_mstrain-608_273e_coco.py
@@ -0,0 +1,121 @@
+_base_ = '../_base_/default_runtime.py'
+# model settings
+model = dict(
+    type='YOLOV3',
+    pretrained='open-mmlab://darknet53',
+    backbone=dict(type='Darknet', depth=53, out_indices=(3, 4, 5)),
+    neck=dict(
+        type='YOLOV3Neck',
+        num_scales=3,
+        in_channels=[1024, 512, 256],
+        out_channels=[512, 256, 128]),
+    bbox_head=dict(
+        type='YOLOV3Head',
+        num_classes=80,
+        in_channels=[512, 256, 128],
+        out_channels=[1024, 512, 256],
+        anchor_generator=dict(
+            type='YOLOAnchorGenerator',
+            base_sizes=[[(116, 90), (156, 198), (373, 326)],
+                        [(30, 61), (62, 45), (59, 119)],
+                        [(10, 13), (16, 30), (33, 23)]],
+            strides=[32, 16, 8]),
+        bbox_coder=dict(type='YOLOBBoxCoder'),
+        featmap_strides=[32, 16, 8],
+        loss_cls=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            loss_weight=1.0,
+            reduction='sum'),
+        loss_conf=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            loss_weight=1.0,
+            reduction='sum'),
+        loss_xy=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            loss_weight=2.0,
+            reduction='sum'),
+        loss_wh=dict(type='MSELoss', loss_weight=2.0, reduction='sum')))
+# training and testing settings
+train_cfg = dict(
+    assigner=dict(
+        type='GridAssigner', pos_iou_thr=0.5, neg_iou_thr=0.5, min_pos_iou=0))
+test_cfg = dict(
+    nms_pre=1000,
+    min_bbox_size=0,
+    score_thr=0.05,
+    conf_thr=0.005,
+    nms=dict(type='nms', iou_thr=0.45),
+    max_per_img=100)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(mean=[0, 0, 0], std=[255., 255., 255.], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile', to_float32=True),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Expand',
+        mean=img_norm_cfg['mean'],
+        to_rgb=img_norm_cfg['to_rgb'],
+        ratio_range=(1, 2)),
+    dict(
+        type='MinIoURandomCrop',
+        min_ious=(0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
+        min_crop_size=0.3),
+    dict(type='Resize', img_scale=[(320, 320), (608, 608)], keep_ratio=True),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(608, 608),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='Pad', size_divisor=32),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=8,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        pipeline=test_pipeline))
+# optimizer
+optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0005)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=2000,  # same as burn-in in darknet
+    warmup_ratio=0.1,
+    step=[218, 246])
+# runtime settings
+total_epochs = 273
+evaluation = dict(interval=1, metric=['bbox'])
diff --git a/docs/model_zoo.md b/docs/model_zoo.md
@@ -147,6 +147,9 @@ Please refer to [Generalized Focal Loss](https://github.com/open-mmlab/mmdetecti
 ### CornerNet
 Please refer to [CornerNet](https://github.com/open-mmlab/mmdetection/blob/master/configs/cornernet) for details.
 
+### YOLOv3
+Please refer to [YOLOv3](https://github.com/open-mmlab/mmdetection/blob/master/configs/yolo) for details.
+
 ### Other datasets
 
 We also benchmark some methods on [PASCAL VOC](https://github.com/open-mmlab/mmdetection/blob/master/configs/pascal_voc), [Cityscapes](https://github.com/open-mmlab/mmdetection/blob/master/configs/cityscapes) and [WIDER FACE](https://github.com/open-mmlab/mmdetection/blob/master/configs/wider_face).

diff --git a/mmdet/core/anchor/__init__.py b/mmdet/core/anchor/__init__.py
@@ -1,10 +1,11 @@
-from .anchor_generator import AnchorGenerator, LegacyAnchorGenerator
+from .anchor_generator import (AnchorGenerator, LegacyAnchorGenerator,
+                               YOLOAnchorGenerator)
 from .builder import ANCHOR_GENERATORS, build_anchor_generator
 from .point_generator import PointGenerator
 from .utils import anchor_inside_flags, calc_region, images_to_levels
 
 __all__ = [
     'AnchorGenerator', 'LegacyAnchorGenerator', 'anchor_inside_flags',
     'PointGenerator', 'images_to_levels', 'calc_region',
-    'build_anchor_generator', 'ANCHOR_GENERATORS'
+    'build_anchor_generator', 'ANCHOR_GENERATORS', 'YOLOAnchorGenerator'
 ]