training error

[xiaoqidaov2]

Encountered while trying to replace Festragin with Festragin of Sinktark

(GaussianTalker) (base) root@53b86a878a7e:~/GaussianTalker# python train.py -s data/Obama/ --model_path output/Obama --configs arguments/64_dim_1_transformer.py Optimizing output/Obama Output folder: output/Obama [23/08 18:31:53] feature_dim: 64 [23/08 18:31:53] [INFO] load aud_features: torch.Size([7999, 29, 16]) [23/08 18:31:53] Reading Training Transforms [23/08 18:31:54] Reading Test Transforms [23/08 18:31:58] Generating Video Transforms [23/08 18:31:59] Loading Training Cameras [23/08 18:31:59] Loading Test Cameras [23/08 18:31:59] Loading Video Cameras [23/08 18:31:59] Deformation Net Set aabb 0.033147845 0.0065655992 [23/08 18:31:59] Voxel Plane: set aabb= Parameter containing: tensor([0.0331, 0.0066]) [23/08 18:31:59] Number of points at initialisation : 3 [23/08 18:31:59] ['xyz', 'deformation', 'grid', 'f_dc', 'f_rest', 'opacity', 'scaling', 'rotation'] [23/08 18:31:59] Training progress: 0%| | 0/7999 [00:00<?, ?it/s]data loading done [23/08 18:31:59] Traceback (most recent call last): File "train.py", line 408, in training(lp.extract(args), hp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from, args.expname, args.use_wandb) File "train.py", line 276, in training scene_reconstruction(dataset, opt, hyper, pipe, testing_iterations, saving_iterations, File "train.py", line 178, in scene_reconstruction loss.backward() File "/root/anaconda3/envs/GaussianTalker/lib/python3.8/site-packages/torch/_tensor.py", line 488, in backward torch.autograd.backward( File "/root/anaconda3/envs/GaussianTalker/lib/python3.8/site-packages/torch/autograd/init.py", line 197, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass RuntimeError: Function _RasterizeGaussiansBackward returned an invalid gradient at index 5 - got [1, 3] but expected shape compatible with [3, 3] Training progress: 0%| | 0/7999 [00:02<?, ?it/s]