search

xyupeng / ContrastiveCrop

[CVPR 2022 Oral] Crafting Better Contrastive Views for Siamese Representation Learning
MIT License
284 stars 27 forks source link

How can I use it for downstream tasks #6

Closed 123456789-qwer closed 2 years ago

123456789-qwer commented 2 years ago

If I want to use representation information for downstream tasks, which part of the feature should I use ？

xyupeng commented 2 years ago

For linear classification, please follow README. For object detection and instance segmentation, we use the same code as in MoCo.

123456789-qwer commented 2 years ago

I still haven't found how the pre-training model is used for target detection, could you give a specific explanation

xyupeng commented 2 years ago

Transferring to detection is the same as MoCo.

xyupeng commented 2 years ago

To be more specific, you need to first convert the saved model (last.pth) to detectron2's format. Then use detectron2's code to train detection.