-
数据分析:data_analysis.ipynb
-
获取test的json:get_testjson.py
-
Retinex与加速测试代码:Retinex.py
已集成在mmdet的transform里,此处仅为测试用
-
WBF: Weighted-Boxes-Fusion/ensemble.ipynb
-
泊松融合:poisson-image-editing和poisson_blending.ipynb
-
画标注框到训练集图上:draw_bbox.ipynb
-
画预测框到测试集图上:draw_pred_bbox.ipynb
-
实例平衡增强:instance_balanced_augmentation.ipynb
-
动态模糊可视化样本图代码:Motion_Blurring.ipynb
-
动态模糊,Mixup和Retinex加入mmdetection后的代码:mmdet/datasets/pipelines/transforms.py,使用时请自行在__init__.py下加入它们的名称。
-
配置文件按照需求修改以下内容:
train_pipeline = [ dict(type='LoadImageFromFile'), dict(type='LoadAnnotations', with_bbox=True), dict(type='Mixup', prob=0.5, lambd=0.8, mixup=True, json_path='data/seacoco/train_waterweeds.json', img_path='data/seacoco/train/'), dict(type='MotionBlur', p=0.3), dict(type='Resize', img_scale=[(4096, 600), (4096, 1000)], multiscale_mode='range', keep_ratio=True), dict(type='Retinex', model='MSR', sigma=[30, 150, 300], restore_factor=2.0, color_gain=6.0, gain=128.0, offset=128.0), dict(type='RandomFlip', flip_ratio=0.5), dict(type='Pad', size_divisor=32), dict(type='Albu', transforms=albu_train_transforms, bbox_params=dict(type='BboxParams', format='pascal_voc', label_fields=['gt_labels'], min_visibility=0.0, filter_lost_elements=True), keymap={ 'img': 'image', 'gt_bboxes': 'bboxes' }, update_pad_shape=False, skip_img_without_anno=True), dict(type='Normalize', **img_norm_cfg), dict(type='DefaultFormatBundle'), dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']), ]
-
标签平滑:mmdet/losses/cross_entropy_loss.py加入了标签平滑的方法。在配置文件需要修改(rpn_head下的CEloss的平滑指数为0.0,bbox_head下的CEloss的平滑指数>0.0即可)如下:
oss_cls=dict( type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0, smoothing=0.001)