NEED HELP: Action Classification low performance

[JayMay1994]

Problems: Use demo to test action classification on kinetics-700 validation set but get very poor result

Experiment:

1. Pretrained model: https://huggingface.co/OpenGVLab/InternVideo2-Stage2_1B-224p-f4/tree/main
2. text candidate：use the class name of k700 dataset annotation
3. dataset：kinetics-700 validation set.
4. code：demo.ipynb

example：
input：carving ice/nTnAoTQ41Nc_000011_000021.mp4
output：
text: coloring in ~ prob: 0.0085
text: acting in play ~ prob: 0.0053
text: smashing ~ prob: 0.0043
text: cracking knuckles ~ prob: 0.0041
text: tasting food ~ prob: 0.0040

[Andy1621]

Can you try the VideoCLIP model?

[JayMay1994]

Can you try the VideoCLIP model?

Do you mean InternVideo2_CLIP in InternVideo2/multi_modality/scripts/evaluation/clip/zero_shot/1B/config_k400.py ？
I'm a green hand in this field :PP

[JayMay1994]

Can you try the VideoCLIP model?

Do you mean abandon internvideo2 and turn to videoclip for action recognition？Your work seems very promising.

[Andy1621]

Yes. I'm not sure whether there is a bug in the demo. But I have tested the VideoCLIP on Kinetics and it runs normally.