Error in EfficientNet V2 MBConvBlock

[sebastian-sz]

System information.

• Have I written custom code (as opposed to using a stock example script provided in Keras): No.
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): N/A
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): 2.11
• Python version: 3.9
• Bazel version (if compiling from source): N/A
• GPU model and memory: N/A
• Exact command to reproduce: N/A

Describe the problem. There is an error in EfficientNetV2's MBConvBlock - the final conv layers are not applied if Squeeze and Excitation block is not applied. This is an error, found by @DavidLandup0 in Keras-CV repo.

"Luckily" for MBConvBlock the SE Block is always applied so no need to change the weights.

Describe the current behavior. Currently, the model works by accident.

Describe the expected behavior. The MBConvBlock should work correctly even without SE Block.

Contributing.

• Do you want to contribute a PR? (yes/no): yes I can.
• If yes, please read this page for instructions
• Briefly describe your candidate solution(if contributing): Basically, we need to unindent lines 717-740 in efficientnet_v2.py

Standalone code to reproduce the issue. N/A

Source code / logs. Current faulty MBConvBlock implementation. Current FusedMBConvBlock implementation, without this issue. David's fixed implementation.

Thanks again @DavidLandup0 for finding this issue.

[tilakrayal]

@sebastian-sz, We would prefer a PR for the validation, Could you please feel free to submit a PR for the requested change to be made. Thank you!

[sebastian-sz]

@tilakrayal PR present at https://github.com/keras-team/keras/pull/17364

[tilakrayal]

The pr has been assigned for reviewing and once it is merged this issue will move to closed status.