Flame
Flame
可以加个联系方式一起探讨下吗，以及一些环境配置问题
How many days you spend to train the roma model? I also find if I replace the dino with other vit the training result is bad
Yes, what I mean is why you choose 16 8 4 2 1 as the scale repeatedly in a same conv_refiner, why not just choose a network which could accept...
Thank you. So, you need a different input channels setting for different resolutions. Do you think is there a powerful network which could use just one parameter setting to do...
When we do the benchmarking, the AUC5, AUC10, and AUC20 values of the original RoMa are 0.622, 0.764, and 0.862, respectively. But, in the paper, they recorded as 62.6, 76.7,...
Hi author, thank you for the kindly answering. Do you get this result from the setting of eval_roma_outdoor? I get result lower 5% than the paper. 
Do you use eval_roma_outdoor.py for paper results? I am not sure why results are {"auc_5": 0.6221797024846101, "auc_10": 0.7651434566786843, "auc_20": 0.8623855457790823, "map_5": 0.8589333333333333, "map_10": 0.8977333333333333, "map_20": 0.9329333333333334} or {"auc_5": 0.6221797024846101, "auc_10":...
Does the megadepth training dataset has some missing files in Undistorted_SfM? Since some of them just have sceneinfo and no matching Undistorted_SfM. When I set k
Yes, I find I turn off autocast in matcher.py and set torch.float32 when define the model could solve this annoying error.
I mean you should delete "with autocast" before the code block in matcher.py