fix(applications): Improve validation and error handling for ConvNeXt weights and fix broadcasting in EfficientNetV2

[harshaljanjani]

• Validate architecture and weights compatibility before API request
• Enhance error messages for mismatched model name and weights
• Fixes A local file was found, but it seems to be incomplete or outdated because the auto file hash does not match the original value of ... #20271

Original Error Message (Unnecessary Model Weights Download):

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/convnext/convnext_large_notop.h5
785596384/785596384 ...
ValueError: Incomplete or corrupted file detected. The auto file hash does not match the provided value of 6fc8009faa2f00c1c1dfce59feea9b0745eb260a7dd11bee65c8e20843da6eab.

New Error Message (Prevents Unnecessary Download and Prompts Early):

ValueError: Model name "convnext_large" does not match weights variant "convnext_small". When using imagenet weights, the model name must contain the weights variant (e.g., "convnext_small").

Unintended Fix For EfficientNetV2:

Fixed a latent bug in EfficientNetV2's preprocessing for channels_first data format where mean and variance tensors were incorrectly shaped as [1,3] instead of [1,3,1,1]. This caused broadcasting errors when normalizing inputs in channels_first mode.

The issue was previously undetected because most users use channels_last format by default. The fix reshapes the normalization parameters to [1,3,1,1] for proper broadcasting across spatial dimensions when using channels_first, while maintaining the original behavior for channels_last format.

Bug revealed by: test_application_*_channels_first test failures
Impact: Affects all EfficientNetV2 models when used with channels_first data format

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 63.63636% with 4 lines in your changes missing coverage. Please review.

Project coverage is 84.52%. Comparing base (a25881c) to head (74e3934).

Files with missing lines	Patch %	Lines
keras/src/applications/convnext.py	33.33%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #20785      +/-   ##
==========================================
+ Coverage   82.01%   84.52%   +2.50%     
==========================================
  Files         555      555              
  Lines       51856    51867      +11     
  Branches     8021     8025       +4     
==========================================
+ Hits        42530    43840    +1310     
+ Misses       7377     5927    -1450     
- Partials     1949     2100     +151     

Flag	Coverage Δ
keras	84.35% <63.63%> (+2.51%)	⬆️
keras-jax	66.63% <63.63%> (+2.52%)	⬆️
keras-numpy	59.02% <0.00%> (+0.11%)	⬆️
keras-openvino	30.04% <0.00%> (+0.13%)	⬆️
keras-tensorflow	67.27% <63.63%> (+2.52%)	⬆️
keras-torch	66.69% <63.63%> (+2.51%)	⬆️
keras.applications	80.24% <63.63%> (?)
keras.applications-jax	80.24% <63.63%> (?)
keras.applications-numpy	22.82% <0.00%> (?)
keras.applications-openvino	22.82% <0.00%> (?)
keras.applications-tensorflow	80.24% <63.63%> (?)
keras.applications-torch	79.98% <63.63%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fchollet]

LGTM, thank you