Something wrong when trying my own dataset

[hanhantie233]

File "train.py", line 372, in <module> training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from) File "train.py", line 41, in training scene = Scene(dataset, gaussians) File "D:\01ZPS_data\models\GaussianPro\scene\__init__.py", line 74, in __init__ self.train_cameras[resolution_scale] = cameraList_from_camInfos(scene_info.train_cameras, resolution_scale, args) File "D:\01ZPS_data\models\GaussianPro\utils\camera_utils.py", line 77, in cameraList_from_camInfos camera_list.append(loadCam(args, id, c, resolution_scale)) File "D:\01ZPS_data\models\GaussianPro\utils\camera_utils.py", line 71, in loadCam sky_mask=resized_sky_mask, normal=resized_normal, depth=resized_depth) File "D:\01ZPS_data\models\GaussianPro\scene\cameras.py", line 53, in __init__ [0, K[1], K[3]], IndexError: index 3 is out of bounds for axis 0 with size 3

I encountered this error when using my own dataset and I'm wondering why.

[hanhantie233]

I found the shape of K is 3.

[kcheng1021]

Please ensure that your dataset format is consistent with the format processed by colmap.

[JohnnyKimura]

@hanhantie233 same problem here，i strictly used colmap to process my dataset and generate the sparse floder and mask floder，but it still had the same issue like you.
it is been a while since you ask the problem，wonder if you have solve it or not，any suggestion？ Thank you！

if i want to process the dataset without mask，what should i do？i simply just delete “--sky_seg” in waymo.sh, it went wrong. but if i provide mask , it have the same problem above... could u tell me how to run the code with the dataset without mask? Thanks again for your amazing work ! @kcheng1021

[hanhantie233]

@hanhantie233 same problem here，i strictly used colmap to process my dataset and generate the sparse floder and mask floder，but it still had the same issue like you. it is been a while since you ask the problem，wonder if you have solve it or not，any suggestion？ Thank you！

if i want to process the dataset without mask，what should i do？i simply just delete “--sky_seg” in waymo.sh, it went wrong. but if i provide mask , it have the same problem above... could u tell me how to run the code with the dataset without mask? Thanks again for your amazing work ! @kcheng1021

I remember setting the camera model in COLMAP to 'PINHOLE' and removing '--sky_seg' to solve my problem.

[JohnnyKimura]

that truly works, thanks!!! but i try to run my indoor dataset it got bad result.

wonder it might cause by the lack of nums of the input images?(my dataset just have 14 images..) or it might cause by the lack of use of the loss function(without sky_seg it may not use some part of the loss function)
and i also want to know any good indoor dataset you can recommend？
@hanhantie233

[hanhantie233]

that truly works, thanks!!! but i try to run my indoor dataset it got bad result.

wonder it might cause by the lack of nums of the input images?(my dataset just have 14 images..) or it might cause by the lack of use of the loss function(without sky_seg it may not use some part of the loss function) and i also want to know any good indoor dataset you can recommend？ @hanhantie233

I believe the most likely reason is that the camera mode you used to capture your dataset was not "PINHOLE." You can check this in COLMAP. Generate sparse point clouds using both "OPENCV" and "PINHOLE" modes and compare their results. If the "PINHOLE" result is poor, it indicates that camera distortion was not disabled when capturing the dataset.
You can also try to increase the number of your dataset, this could also be a potential reason.

[JohnnyKimura]

that truly works, thanks!!! but i try to run my indoor dataset it got bad result.
wonder it might cause by the lack of nums of the input images?(my dataset just have 14 images..) or it might cause by the lack of use of the loss function(without sky_seg it may not use some part of the loss function) and i also want to know any good indoor dataset you can recommend？ @hanhantie233

I believe the most likely reason is that the camera mode you used to capture your dataset was not "PINHOLE." You can check this in COLMAP. Generate sparse point clouds using both "OPENCV" and "PINHOLE" modes and compare their results. If the "PINHOLE" result is poor, it indicates that camera distortion was not disabled when capturing the dataset. You can also try to increase the number of your dataset, this could also be a potential reason.

that really help me，thanks a lot my friend！ @hanhantie233