test on dynamic scene dataset used by Neural Scene Flow Fields (NSFF)....

[sfchng]

Dear authors,

Thanks for open sourcing the code. I wonder if you guys have tested it on the dynamic scene sequences released by Nvidia (https://drive.google.com/drive/folders/1G-NFZKEA8KSWojUKecpJPVoq5XCjBLOV) before? I tried to write my own dataloader (by adapting the readCamerasFromNpy function in dataset_readers.py , and tested it on the Truck-2 sequences, but I did not manage to get it train properly with both control-point-based deformation and MLP-based deformation. For your information, I am not providing the algorithm with any mask information.

Here is my run my command. Could it be that I may have missed some essential configurations or is it due to the noisy camera pose estimations? python train_gui.py --source_path="./data/nvidia/Truck-2" --model_path=outputs/nvidia-truck8 --node_num=2048 --deform_type="node" --gui --init_isotropic_gs_with_all_colmap_pcl

python train_gui.py --source_path="./data/nvidia/Truck-2" --model_path=outputs/nvidia-truck8 --deform_type="mlp" --gui --init_isotropic_gs_with_all_colmap_pcl

Cheers