Image channel size is lost after resizing layer.

[linus87]

Please make sure that this is a bug. As per our
GitHub Policy,
we only address code/doc bugs, performance issues, feature requests and
build/installation issues on GitHub. tag:bug_template

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow.js): Yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): MAC 14.2.1 (23C71)
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:
• TensorFlow.js installed from (npm or script link): script link
• TensorFlow.js version (use command below): 4.20.0
• Browser version: Chrome 126.0.6478.127
• Tensorflow.js Converter Version:

Describe the current behavior
Channel dimension size is lost.

Describe the expected behavior
Image is resized, but the channel is the same as input.

Standalone code to reproduce the issue
Provide a reproducible test case that is the bare minimum necessary to generate
the problem. If possible, please share a link to Colab/CodePen/any notebook.
`
const model = tf.sequential();
const imageSize = 200;

model.add(tf.layers.inputLayer({batchInputShape: [1, null, null, 3]}));
model.add(tf.layers.resizing({height: imageSize, width: imageSize, interpolation: 'bilinear', cropToAspectRatio: true}));
model.summary();
`

Other info / logs Include any logs or source code that would be helpful to
diagnose the problem. If including tracebacks, please include the full
traceback. Large logs and files should be attached.

[gaikwadrahul8]

Hi, @linus87

Thank you for bringing this issue to our attention and I'm able to replicate the same behavior from my end which you reported in the issue template so we'll have to dig more into this issue and will update you soon, for reference I've added screenshot below :

Thank you for your cooperation and patience.

[github-actions]

This issue has been marked stale because it has no recent activity since 7 days. It will be closed if no further activity occurs. Thank you.

[gaikwadrahul8]

Hi, @linus87

I apologize for the delay in my response and as far I know, the image channel size (number of color channels) is not actually lost after the resizing layer in your TensorFlow.js model, even though model.summary() might show null for the output shape's channel dimension in the resizing layer because model.summary() provides a high-level overview of the model architecture. It makes some inferences based on the layer configuration but might not explicitly show all details.

In your code, the resizing layer tf.layers.resizing doesn't perform any operations that would remove color channels. It's designed to resize the image dimensions (height and width) while maintaining the existing number of channels.

The null in the output shape [200, 200, null] for the resizing layer indicates that the exact number of channels is still unknown at this point in the model summary. This is because the input layer (input1) accepts images with an unknown height and width (null) and consequently, the number of channels might also be unknown until an actual image is fed into the model.

To verify that the channel information is preserved, you can try the following:

1. Define a sample image_data with known dimensions and channels
2. Pass the sample image_data through the model and do predictions with the help of model.predict(image_data)
3. Examine the shape of the output returned by model.predict(). It should maintain the original number of color channels(3) and The model summary might show null for the channel dimension in the resizing layer's output shape. However, the output shape of the prediction should be [1, 200, 200, 3]

This will confirm that even though model.summary() shows null for channels in the resizing layer's output shape but actual processing preserves the channel information during resizing.

To confirm the preservation of no. of color channels, could you please run the model and verify if the output shape from model.predict() maintains the original number of channels (3) in your example ?

If I have missed something here please let me know.

Thank you for your cooperation and patience.