Question about technical details - Githubissues

SHI-Labs / Decoupled-Classification-Refinement

Revisiting RCNN: On Awakening the Classification Power of Faster RCNN (ECCV 2018)

https://arxiv.org/abs/1803.06799

MIT License

167 stars 24 forks source link

Question about technical details #15

Open hdjang opened 5 years ago

hdjang commented 5 years ago

Hi, thanks for sharing your great work.

I have a question regarding technical details of your works.

In Table.1 in your paper(ECCV2018), you compare 'sampling methods' to choose training samples of DCR module. Can you let me know 'precise' pool of samples in each sampling method? It looks a little confusing for me since 'FP only (hard false positive only)' can also include foreground(FG) or background(BG) samples. Did you mean the sampling pool of each sampling method as pool that 'intersects' all cases(e.g., FP, FG, and BG)?

bowenc0221 commented 5 years ago

Please refer to these codes: https://github.com/bowenc0221/Decoupled-Classification-Refinement/blob/dcr_v1/dcr_v1/core/rcnn.py#L73-L93 for the exact sampling method.