Add DeepLab example #1534
Add DeepLab example #1534
Conversation
|
Adding Ian as reviewer and Jon for visibility. |
| """ | ||
| Title: Semantic segmentation using DeepLabV3+ | ||
| Author: [Alexia Audevart](https://github.com/aaudevart) | ||
| Date created: 2023/13/04 |
There was a problem hiding this comment.
Is this meant to be September 13? (13/09)?
Either way -- YYYY/MM/DD format is the standard here so let's use that.
|
|
||
| """shell | ||
| !pip install keras-core --upgrade | ||
| !pip uninstall -y keras-cv |
There was a problem hiding this comment.
I don't think we should need this uninstall for the example
| Detect hardware, return appropriate distribution strategy | ||
| """ | ||
|
|
||
| try: |
There was a problem hiding this comment.
For the example, I think we can exclude this bit and forego a strategy altogether. TF will just default to using the GPU in this case so we don't need to worry about distribution
| train_gen = tf.data.Dataset.from_generator( | ||
| generator=gen_pairs(train_img_list), | ||
| output_signature=( | ||
| tf.TensorSpec(shape=(IMG_WIDTH, IMG_HEIGHT, 3), dtype=tf.uint8), |
There was a problem hiding this comment.
Not critical here, since images are square, but this should be IMG_HEIGHT, IMG_WIDTH (order reversed)
| val_gen = tf.data.Dataset.from_generator( | ||
| generator=gen_pairs(val_img_list), | ||
| output_signature=( | ||
| tf.TensorSpec(shape=(IMG_WIDTH, IMG_HEIGHT, 3), dtype=tf.uint8), |
There was a problem hiding this comment.
(Same here re: height,width)
| Adam as the optimizer. | ||
| """ | ||
|
|
||
| EPOCHS = 16 |
There was a problem hiding this comment.
For examples, we often try to keep the number of epochs low -- how long does this take to train?
If it's possible to get decent results with fewer epochs we should do that.
divyashreepathihalli
requested review from
divyashreepathihalli
and removed request for
jbischof
Add a new example using DeepLabV3+ pretrained model for semantic segmentation in healthcare
Google colab : https://colab.research.google.com/drive/17UQr7-UM_Et0A7g7638Gq7Vb-e1171kC?usp=sharing