metadata

license: mit
datasets:
  - mjjung/ActivityNet-VTune
language:
  - en

TimeChat-7B-ActivityNet-VTune Model

Model details

We trained TimeChat using VTune, a developed instruction-tuning method specifically designed to account for consistency.

For the tuning, we utilized 10K training videos from ActivityNet-Captions with 205K automatically generated annotations.

Evaluation

We evaluated the model on ActivtyNet-CON and ActivtyNet-Captions.

ActivityNet-CON

Metric Value

Ground 37.4

R-Ground 28.3 (75.6)

S-Ground 10.6 (28.3)

H-Verify 19.6 (52.3)

C-Verify 19.5 (51.5)
ActivityNet-Captions

Metric Value

R@1 IoU=0.3 57.74

R@1 IoU=0.5 41.05

R@1 IoU=0.7 23.72

mIoU 40.89

Metric	Value
Ground	37.4
R-Ground	28.3 (75.6)
S-Ground	10.6 (28.3)
H-Verify	19.6 (52.3)
C-Verify	19.5 (51.5)

Metric	Value
R@1 IoU=0.3	57.74
R@1 IoU=0.5	41.05
R@1 IoU=0.7	23.72
mIoU	40.89

Paper and Code for more information: Paper, Code

Citation

If you find our research and codes useful, please consider starring our repository and citing our paper:

@article{jung2024consistency,
  title={On the Consistency of Video Large Language Models in Temporal Comprehension},
  author={Jung, Minjoon and Xiao, Junbin and Zhang, Byoung-Tak and Yao, Angela},
  journal={arXiv preprint arXiv:2411.12951},
  year={2024}
}