This includes my code for Tianchi competition: Alum surface defect detection. (held by Alibaba company)

The competition is aimed at using computer vision techniques to help workers check whether their Aluminum surface products have any defects such as spots, scratches and so on.

• InceptionV4(pytorch)
• combine vote(similar to bagging)
• good iteration steps

• data augmentation(horizontal flip)
• Ensemble(Xception, Resnet50, InveptionV3)

• All other augmentation. especially random rotation(0~8, the larger angle, the worse acc))

• FasterRcnn&FPN(detectron)
• Larger resize size(960 for maskrcnn, 800 for FasterRcnn)
• bbox vote
• Adam instead of SGD
• lower the thresh

• Mask-Rcnn(keras tf)
• YoloV3
• FasterRcnn(tf)
• Emsemble(FPN, faster-rcnn, mask-rcnn)
• Soft-nms(since few defects have overlap)
• bbox-vote strategy(ID, AVG, IOU_AVG)
• Delete mini batch(since spots are super small)
• Data augmentation( train&test scales, flip, small rotation)

• my own bbox vote( similar to softer-nms, a combination of iou and confidence)
• bbox combination ( similar to this from kaggle big bowl 2018)
• Use larger size(1920x2560), more data augmentation(5 scales etc.)...

• SNIPER
• Cascade-rcnn
• maskrcnn(X152 backbone)

• Official discussion forum
• DCN github link
• 1st place's article sharing