Movinet kinetics 400 weights are actually kinetics 600 weights

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• [* ] I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• [ *] I am reporting the issue to the correct repository. (Model Garden official or research directory)
• [ *] I checked to make sure that this issue has not been filed already.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/master/official/projects/movinet#kinetics-400

2. Describe the bug

The published checkpoint weights for Kinetics-400 are identical to those for Kinetics-600. For example, the following checkpoint weights under Kinetics-600:
https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a0_base.tar.gz
are identical to the following checkpoint weights under kinetics-400:
https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a0_base_k400.tar.gz

3. Steps to reproduce

Download and extract both checkpoints:

wget https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a0_base.tar.gz
tar --one-top-level=600 -xvf movinet_a0_base.tar.gz
wget https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a0_base_k400.tar.gz
tar --one-top-level=400 -xvf movinet_a0_base_k400.tar.gz

Compare checkpoints and observe they are equal:
diff 600/movinet_a0_base/ckpt-1.data-00000-of-00001 400/movinet_a0_base/ckpt-1.data-00000-of-00001

4. Expected behavior

The published Kinetics-400 weights should correspond to those trained on Kinetics-400. It is likely that the Kinetics-600 weights were mistakenly uploaded instead of the Kinetics-400 weights. To clarify, loading the published Kinetics-400 checkpoint weights into official.projects.movinet.modeling.movinet_model.MovinetClassifier with num_classes=400 results in an error.

5. Additional context

6. System information

OS Platform and Distribution - Ubuntu 22.04.5 LTS
TensorFlow installed from (source or binary): binary
TensorFlow version (use command below):=2.18.1
Python version: 3.10.12
CUDA/cuDNN version: cuda_12.8.r12.8
GPU model and memory: NVIDIA GeForce RTX 4090, 24GB

Start with the MoViNet Kinetics-400 documentation at official/projects/movinet and reproduce the reported comparison using the two checkpoint URLs and the provided wget, tar, and diff commands. Check how the checkpoint is loaded by official.projects.movinet.modeling.movinet_model.MovinetClassifier with num_classes=400; done means the published Kinetics-400 archive contains distinct weights that load successfully.