search

jshtok / RepMet

Few-shot detection for visual categories
Apache License 2.0
108 stars 18 forks source link

finetune with few shot classes #7

Closed liyangliu closed 4 years ago

liyangliu commented 4 years ago

image The right part of Tab. 3 shows the results of episode fine tuning, and the paper said, image

Can you explain how you actually fine tune with the episodic data, I am not so sure what is the last layer of the network? Thanks.

jshtok commented 4 years ago

Dear Liyang,

There is a 'freeze' parameter in the code, which admits values 0 (all layers are training), 1 and 2 (most layers are frozen. freeze=2 leaves the last two fully connected layers and the bounding box prediction layer unfrozen, and freeze=1 leaves only the representatives unfrozen. We did the finetuning with the freeze=1 option.