Open bulutthecat opened 10 months ago
+1. Also is there a plan to release the inference code? Thanks.
followed
Hi,
just to clarify, this method is a test-time optimization-based method (i.e., it needs to be optimized separately on each video), so there is not a single set of pre-trained weights and biases. However, we will try to release the pretrained weights for a few sequences, to help people who want to see what the results/formats look like without having to run optimization.
As for the inference code, we are basically following the evaluation protocol of the TAP-Vid benchmark, and we added one more metric of temporal coherence. We plan to release our inference code soon.
Are there any plans on releasing the pre-trained weights and biases of this model?