Course Study with CSC321
[2021.04.10]
Topic
: Lecture 13.4: Object Detection : Fast R-CNN details, ResNet fine-tuning practice (James , Jaden)
Notes
:
- https://drive.google.com/file/d/1ari0YxYTqaH9mky2pKkPOO9p15gX1EBS/view?usp=sharing [James]
- https://drive.google.com/drive/folders/18WZeNJSrlOti07epXNy75Ws1s6U-Y-Q9?usp=sharing [Jaden]
Links
:
Fast R-CNN
- https://arxiv.org/pdf/1504.08083.pdf (original fast r-cnn thesis)
- https://fintecuriosity-11.tistory.com/73 (ablation study)
- https://yeomko.tistory.com/15 (how is end-to-end training possible?)
- https://deepsense.ai/region-of-interest-pooling-explained/ (spp vs roi pooling)
- https://ratsgo.github.io/deep%20learning/2017/04/05/CNNbackprop/ (backprops in CNN layer)
Res Net fine-tuning code
- https://www.tensorflow.org/hub/tutorials/tf2_object_detection
- https://detectron2.readthedocs.io/en/latest/_modules/detectron2/modeling/roi_heads/fast_rcnn.html
- https://github.com/facebookresearch/detectron2/blob/master/detectron2/modeling/roi_heads/fast_rcnn.py
- https://github.com/rbgirshick/py-faster-rcnn
Covered through study
- Fast R-CNN :
- what’s improved from SPP Net ? (idea, keywords)
- end-to-end
- softmax replacing svm
- ROI-pooling
- Fast R-CNN flow
- comapred with previous models
- training Fast R-CNN
- multi-task loss function
- Hierarchical Sampling ( vs. region-wise sampling)
- test methods (truncated SVD)
- Fast R-CNN Limits
- ResNet fine-tuning source code
Fast R-CNN overview & details