- 作者: 廖满文、吴雨暄、饶龙、汪潇翔
- 指导老师: 王兴刚、廖振宇
- Object localization using CNN
- Data: http://xinggangw.info/data/tiny_vid.zip
- For every class, the first 150 images are used for training the localizer and the rest for testing.
- Evaluation metric: A correct localization means the predict class label is correct and the predict box has an IoU>0.5 with the ground-truth. You should report the average accuracy and show some localization results.
- Step #1: implementing the localization network using direct bbox regression
- Step #2: adding anchors in the localization network
- Extra credit: multiple object localization using your own network