Reproduction of Tanks and Temples dataset results
phuong-dinhmai opened this issue · comments
Hi all. I am trying to reproduce the results of Tanks and Temples dataset. I am running the tnt_eval.py file in your code, and the results are quite low. For instance, the evaluation metrics for the Meetingroom scene are:
distance tau : 0.010
precision : 0.0319
recall : 0.0301
f-score : 0.0310
SSIM : 0.8645592
PSNR : 25.1977291
LPIPS: 0.2680884
I am not sure what I did wrong when training the model. Does anyone know how to reproduce the results?
P/s: I am using A5000 instead of RTX 3090
Do you inspect on the output PLY files? Is the input file aligned well to the ground truths? Someone also faced similar issue and it turns out that they input wrong trajectory file. Please let me if the issue further exists.