Zhimin00

I downloaded the files in layers directory from https://github.com/huggingface/pytorch-image-models/tree/main/timm/layers and substituted the corresponding directory of 'mm-cot/timm/models/layers', which solved the ImportError.

Hi, @Parskatt, does training batchsize significantly affect the final performance of the model? I only have 24G gpus, and the maximum batchsize is 2 rather than 8 you used as...

Thank you for your fast response. Do you mean a batch size of 8 in total (global batch size) or 8 per GPU (local batch size)?

I see. bn is quite annoying!

Hi, Parskatt. Can you provide detailed codes for creating the matched points pic? Thank you. > Hi, Im away for the summer, will get back with some demos start of...

You can find it in codes related to benchmark, for example the geometric_dist in MegadepthDenseBenchmark returns 'epe' and pcks.

Thanks for your awesome work. Have you test RoMa on _ZEB_ and $$GIM_{RoMa}$$ on HPatches, MegaDepth1500, Scannet1500, Aachen1.1 and Inloc? I was wondering if $$GIM_{RoMa}$$ could be the new sota...

Hi, I have tested on MegaDepth1500, Scannet1500, and HPatches. Here are numbers for your reference. And I think it experiences much more degradation than other models.   | MegaDepth-1500 |   |  ...

I totally get your points and that's one of your contribution. Thank you! Just to clarify, how many image pairs were generated from the 50-hour and 100-hour videos?

I was trying to reproduce the results of RoMa (out) and RoMa (in) on ZEB (using the same pipeline as gim_roma, only replacing the checkpoints with roma_outdoor or roma_indoor from:...