Support for partial encoder freezing #341
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
Allow you to freeze the backbone encoder partially. For example, you can set encoder_freeze=0.8 which will make the first 80% layers untrainable, and leave the last 20% trainable.
This lets you tune the trainability of your network.
I freeze early layers first because: early layers learn basic features that are less likely to change between data sets. Late layers learn abstract features that may change more between data sets.
In my experiment, setting a 95% freeze was good for accuracy AND training speed for Unet + inceptionv3 on the Oxford-IIIT Pet dataset (code at https://github.com/AllWashedOut/pet_segmentation)
0% frozen
python .\model.py --patience=10 --encoder_freeze_percent=0.0
Test accuracy: 90.0%. Epoch time: 60s
95% frozen
python .\model.py --patience=10 --encoder_freeze_percent=0.95
Test accuracy: 91.13%. Epoch time: 52s
100% frozen
python .\model.py --patience=10 --encoder_freeze_percent=1.0
Test accuracy: 90.99%. Epoch time: 50s