“华为云杯”2020深圳开放数据应用创新大赛·生活垃圾图片分类-分享

Details

• Dataaug: edit resize and mirror augment method
• Backbone: se_resnext101_32x4d
• neck: fpn
• loss: ciou-loss

Test result

• weight path(提取码：lnow)
• put the weight in weights/train_weights/ssd300_se_resnext101_32x4d_correct_transform_full_data_ciou_sgd_1gpu_6_6/, than run the lajiEval.py

Results

• map 72.66

Reference

Zzh-tju/DIoU-SSD-pytorch