Training pipeline #26
Conversation
pshivraj
requested review from
havanagrawal,
lmtoan,
tejasmhos and
vivanvish
|
I think the training pipeline is fine. It can be merged to the master branch. |
|
|
||
| def model_train(self): | ||
|
|
||
| augmentation = iaa.Sequential([ |
There was a problem hiding this comment.
Keep this line number and file in mind whenever Yipeng asks if we perform any image augmentation during training.
|
Thanks Puru. Since we need to train images with no bottle/box/bag, should we stash more commit on this branch or merge this and create another feature branch? Just cosmetics:
|
| """ | ||
| Initialize the class with dataset info. | ||
| """ | ||
| with open(ROOT_DIR + 'id_map.json') as mapping_file: |
There was a problem hiding this comment.
Could you check in the id_map.json file? Or I can add that long list of class labels in config.py (with "book" becomes "bags").
There was a problem hiding this comment.
@lmtoan I have added the id_map.json to the pull request. Can you approve my latest commit so that I can merge.
pshivraj
dismissed stale reviews from havanagrawal and tejasmhos
via
619a958
|
I am closing this pull request for now. |
This PR is an end to end pipeline for model Training. It includes:
Data pre-processing using
pre_process.pyto generate .h5 file for masks.Custom Config for clomask Dataset. (
config.py)Model Training using coco and imagenet weights. (
train.py)This caters to following issues: #18 #17 #26