Jianyuan Wang

The issue is closed and please feel free to re-open it.

Hi, The script demo.py and evaluate.py come from the repository https://github.com/princeton-vl/RAFT. They were designed for RAFT visualization and evaluation, but not our method. I just kept them for completeness. I...

Hi, we did not test our method on the ScanNet dataset.

Hi, The flag cfg.PRED_POSE_GT_SCALE was only used for the ablation study `Oracle Scale' in Table 6 of the paper, and not for VO evaluation. For a quick look, here are...

Btw, I was wondering if you can reproduce the depth results, or encounter a similar situation to @longyangqi as in issue #6? Several people have reported that they can reproduce...

It is weird. It seems the compiled RANSAC module of you still does not work correctly. But even the reproduced depth results of you two are very different.... I may...

Hi, Would you mind checking the batch size of the training? Also, it would be helpful if you could print the network outputs layer by layer to find where the...

Hi guys @longyangqi @Zero314159 , It looks like something is wrong in the process of _test.py_ or some corner cases occur. Have you tried to ignore the testing process, and...

> @jytime I can run the evaluation codes without any error. The evaluation is produced according to your reply in #7 , using the 30% semi-dense official ground-truth depth if...

Hi @longyangqi, I checked with the guys who can reproduce the results, which should be close to the figure below.  Regarding the data preparation process, could you please verify...