er-muyue / DeFRCN

MIT License
184 stars 44 forks source link

Experimental setting is different from TFA and FSRW #52

Open 1179021477 opened 2 years ago

1179021477 commented 2 years ago

In TFA/FSRW, FSOD fine-tune on both base and novel classes, using the same training examples provided by FSRW. And G-FSOD train for multiple runs on different random samples to obtain averages. Can you explain this? The results of only fine-tune on novel classes is not fair to compare with others