yuanyunshuang / fpv_rcnn Goto Github PK

Hi, I download your COMAP dataset and extract them all, but find them confusing. The cooperative vehicles in 'tfs' and 'coop_cloud' have no correspondence

Take the first frame in train_val.txt as example, 1070_000317.

keys in tfs/1070_000317.npy include ['tf_ego', '000578',''000573'',''000561'']

files in coop_cloud/1070_000317 include ['000388.bin', '000447.bin', '000386.bin', '000414.bin', '000458.bin', '000387.bin', '000391.bin', '000399.bin', '000405.bin']. And them will turn to coop_ids.

so I encounter error in line 116 in datasets/comap/comap.py, which says a key error.

Greetings,

I can not find any image in data and corresponding ground truth data for semantic segmentations.

Regards,
Melih

hello, I can't download the COMAP dataset. The link https://seafile.cloud.uni-hannover.de/d/1c52826e98d34c0399a4/ interface's shows that '/COMP/' does not exist.

Hi, thanks for your great work! I was wondering, have you tried your method on some other dataset, such as OPV2V? It would be good if you have tried and provide a number on it. Thanks!

Hi! When I try to run tools/train_fusion_detector.py with cfg/fusion_pvrcnn_comap.py, I find I fall in a dead loop in MatcherV3' s clustering method.

cluster_indices.shape: torch.Size([1608])
ious.shape: torch.Size([1608, 1608])
torch.where(ious[cur_idx] > 0.1) returns all empty tensor in the end, like (tensor([], device='cuda:0', dtype=torch.int64),).

Then I simply replace MatcherV3 with MatcherV2, I meet another error:

File "/GPFS/data/yifanlu/FPV_RCNN/losses/smooth_l1_loss.py", line 55, in forward
    anchorwise_smooth_l1norm *= weights.unsqueeze(-1) # pos_anchors multiply the weight: 1/num_pos_anchor in each sample
...
RuntimeError: Function 'MulBackward0' returned nan values in its 0th output.

There are some other questions I look forward to your answers:

I find the pretrained model cia_ssd.pth you provide don't come in handy. Where to load it?

If I do not have the class labels of points, can I train the SemSeg network? Since I want to apply this method in other dataset.