Open bhack opened 5 months ago
Thank you for sharing. We have only tested the commonly used ROVS benchmarks to facilitate comparison with previous methods. Since this is a relatively new dataset, we haven’t tried it yet. We might consider experimenting with it to evaluate our model’s performance once the benchmark is released after May 30th.
This work is interesting especially waiting for Sora level architecture quality reproduction. Have you tried to evaluate/stress test it on the new CVPR 2024 challenge? https://henghuiding.github.io/MeViS/ChallengeCVPR2024