training with preprocessed data failed

[Alen-Wong]

Thank you for this excellent work, I encountered the following error when training with preprocessed data, how can I resolve it?

Optimizing 
Output folder: ./output/3765832a-c
Reading camera 8/8
Loading Training Cameras
[ INFO ] Encountered quite large input images (>1.6K pixels width), rescaling to 1.6K.
 If this is not desired, please explicitly specify '--resolution/-r' as 1
Loading Test Cameras
Number of points at initialisation :  2108
Training progress:   6%|▊            | 600/10000 [01:10<17:07,  9.15it/s, Loss=0.2673802]Traceback (most recent call last):
  File "train.py", line 322, 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 223, in training
    gaussians.densify_and_prune(opt.densify_grad_threshold, 0.005, scene.cameras_extent, size_threshold)
  File "/media/data_nix/wangzihao/code/depth-aware-3DGS/scene/gaussian_model.py", line 393, in densify_and_prune
    self.densify_and_clone(grads, max_grad, extent)
  File "/media/data_nix/wangzihao/code/depth-aware-3DGS/scene/gaussian_model.py", line 384, in densify_and_clone
    new_scaling = self._scaling[selected_pts_mask]
IndexError: The shape of the mask [2108] at index 0 does not match the shape of the indexed tensor [1, 3] at index 0
Training progress:   6%|▊            | 600/10000 [01:10<18:18,  8.56it/s, Loss=0.2673802]

[YoungXusheng]

The calculation of dist2 was wrong,u can check the gaussian_module

[liushuishui7]

Have you solved this problem? I had the same problem。

[liushuishui7]

python train.py -s ./data/fern --scene fern Optimizing&nbsp; Output folder: ./output/647b96f1-4 [17/04 02:51:52] Tensorboard not available: not logging progress [17/04 02:51:52] Checking scene type in path: /data/lmb/depth-aware-3DGS/data/fern [17/04 02:51:52] Subdirectories found: ['database.db', 'sparse', 'images_low_res', 'depth', '.DS_Store', 'images', 'abs_depth'] [17/04 02:51:52] Reading camera 8/8 [17/04 02:51:52] Loading Training Cameras [17/04 02:51:52] [ INFO ] Encountered quite large input images (&gt;1.6K pixels width), rescaling to 1.6K. &nbsp;If this is not desired, please explicitly specify '--resolution/-r' as 1 [17/04 02:51:52] Loading Test Cameras [17/04 02:51:57] Number of points at initialisation :&nbsp; 2108 [17/04 02:51:57] Training progress:&nbsp; &nbsp;6%|███▌&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;| 600/10000 [01:14<18:29,&nbsp; 8.48it/s, Loss=0.2663194]Traceback (most recent call last): &nbsp; File "/data/lmb/depth-aware-3DGS/train.py", line 326, in <module&gt; &nbsp; &nbsp; 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) &nbsp; File "/data/lmb/depth-aware-3DGS/train.py", line 225, in training &nbsp; &nbsp; gaussians.densify_and_prune(opt.densify_grad_threshold, 0.005, scene.cameras_extent, size_threshold) &nbsp; File "/data/lmb/depth-aware-3DGS/scene/gaussian_model.py", line 396, in densify_and_prune &nbsp; &nbsp; self.densify_and_clone(grads, max_grad, extent) &nbsp; File "/data/lmb/depth-aware-3DGS/scene/gaussian_model.py", line 387, in densify_and_clone &nbsp; &nbsp; new_scaling = self._scaling[selected_pts_mask] IndexError: The shape of the mask [2108] at index 0 does not match the shape of the indexed tensor [1, 3] at index 0 Training progress:&nbsp; &nbsp;6%|███▌&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;| 600/10000 [01:14<19:30,&nbsp; 8.03it/s, Loss=0.2663194] 您好我在复现您的论文代码时，遇到了这个问题，向您请教如何处理解决

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "raja-kumar/depth-aware-3DGS" ***@***.***&gt;; 发送时间:&nbsp;2025年4月15日(星期二) 晚上10:55 ***@***.***&gt;; ***@***.******@***.***&gt;; 主题:&nbsp;Re: [raja-kumar/depth-aware-3DGS] training with preprocessed data failed (Issue #3) 您好，我是勇旭升。谢谢您的邮件，我已经收到了。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt; YoungXusheng left a comment (raja-kumar/depth-aware-3DGS#3) 您好，我是勇旭升。谢谢您的邮件，我已经收到了。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[raja-kumar]

@liushuishui7 The diff-gaussian-rasterization module is not linked correctly. can you please install the diff-gaussian-rasterization from this repo https://github.com/ashawkey/diff-gaussian-rasterization and try again

[liushuishui7]

I have tried your solution, but this problem still occurs. The length of selected_pts_mask is inconsistent with the current number of Gaussian points, which may be caused by logical errors in the gradient calculation or pruning steps? (depth_aware_3dgs) ***@***.***:/data/lmb/depth-aware-3DGS$ python train.py -s ./data/fern --scene fern Optimizing&nbsp; Output folder: ./output/1e3fa6bc-8 [19/04 21:29:05] Tensorboard not available: not logging progress [19/04 21:29:05] Checking scene type in path: /data/lmb/depth-aware-3DGS/data/fern [19/04 21:29:05] Subdirectories found: ['database.db', 'sparse', 'images_low_res', 'depth', '.DS_Store', 'images', 'abs_depth'] [19/04 21:29:05] Reading camera 8/8 [19/04 21:29:06] Loading Training Cameras [19/04 21:29:06] [ INFO ] Encountered quite large input images (&gt;1.6K pixels width), rescaling to 1.6K. &nbsp;If this is not desired, please explicitly specify '--resolution/-r' as 1 [19/04 21:29:06] Loading Test Cameras [19/04 21:29:11] Number of points at initialisation :&nbsp; 2108 [19/04 21:29:11] Training progress:&nbsp; &nbsp;6%|███▌&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;| 600/10000 [01:14<18:38,&nbsp; 8.40it/s, Loss=0.2659831]Traceback (most recent call last): &nbsp; File "/data/lmb/depth-aware-3DGS/train.py", line 326, in <module&gt; &nbsp; &nbsp; 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) &nbsp; File "/data/lmb/depth-aware-3DGS/train.py", line 225, in training &nbsp; &nbsp; gaussians.densify_and_prune(opt.densify_grad_threshold, 0.005, scene.cameras_extent, size_threshold) &nbsp; File "/data/lmb/depth-aware-3DGS/scene/gaussian_model.py", line 396, in densify_and_prune &nbsp; &nbsp; self.densify_and_clone(grads, max_grad, extent) &nbsp; File "/data/lmb/depth-aware-3DGS/scene/gaussian_model.py", line 387, in densify_and_clone &nbsp; &nbsp; new_scaling = self._scaling[selected_pts_mask] IndexError: The shape of the mask [2108] at index 0 does not match the shape of the indexed tensor [1, 3] at index 0 Training progress:&nbsp; &nbsp;6%|███▌&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;| 600/10000 [01:14<19:23,&nbsp; 8.08it/s, Loss=0.2659831]

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "raja-kumar/depth-aware-3DGS" ***@***.***&gt;; 发送时间:&nbsp;2025年4月20日(星期天) 上午7:14 ***@***.***&gt;; ***@***.******@***.***&gt;; 主题:&nbsp;Re: [raja-kumar/depth-aware-3DGS] training with preprocessed data failed (Issue #3) @liushuishui7 The diff-gaussian-rasterization module is not linked correctly. can you please install the diff-gaussian-rasterization from this repo https://github.com/ashawkey/diff-gaussian-rasterization and try again — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***&gt; raja-kumar left a comment (raja-kumar/depth-aware-3DGS#3) @liushuishui7 The diff-gaussian-rasterization module is not linked correctly. can you please install the diff-gaussian-rasterization from this repo https://github.com/ashawkey/diff-gaussian-rasterization and try again — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***&gt;

[knight-joker]

I also have this problem ,and spend too much time but filed to solve it

[knight-joker]

Optimizing
Output folder: ./output/c93aaa39-1 [15/04 16:10:13]
Tensorboard not available: not logging progress [15/04 16:10:13]
Reading camera 8/8 [15/04 16:10:13]
Loading Training Cameras [15/04 16:10:13]
Loading Test Cameras [15/04 16:10:14]
Number of points at initialisation : 2108 [15/04 16:10:14]
Training progress: 0%| | 0/10000 [00:00<?, ?it/s]Traceback (most recent call last):
File "C:\Users\lenovo\Desktop\3Drebuild\depth-aware-3DGS\depth-aware-3DGS\train.py", line 364, in
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 "C:\Users\lenovo\Desktop\3Drebuild\depth-aware-3DGS\depth-aware-3DGS\train.py", line 114, in training
assert gaussians._scaling.shape[0] == n_pts,
AssertionError: [FATAL] _scaling shape mismatch: 1 vs xyz 2108. Check prune/densify logic.
Training progress: 0%| | 0/10000 [00:00<?, ?it/s]

[knight-joker]

I give up this