Donny Chen comments

Results 80 comments of


                                            Donny Chen

Question about Camera Conventions

Hi, @YeZhang0621, thanks for your appreciation. Sorry for the confusion caused. The reason why we inverse it is because we assumed that `w2c` is stored in the torch file in...

excuse me， I used the pre packaged model provided in readme for inference without making any changes to the code, and the result I obtained was fuzzy without any changes to the code. What could be the reason for this?

Hi @songofjoy, would you mind sharing some of your results here? MVSplat360 is, of course, not a perfect solution for every scene, but the released model should be able to...

excuse me， I used the pre packaged model provided in readme for inference without making any changes to the code, and the result I obtained was fuzzy without any changes to the code. What could be the reason for this?

Hi @songofjoy, I just rerun the test on my machine using the latest released codes. Below are the outputs I obtained regarding the two images you shared. ![Screenshot 2024-12-26 at...

Kindly Inquiry for Evaluation Issue with Self-Converted DL3DV Chunks in MVSplat360

Hey @YaoXingbo, you might want to look into the DL3DV data loader, which relies on `self.chunks`(chunk path) and `self.index`(scene index) to fetch the testing item. In particular, you can uncomment...

Kindly Inquiry for Evaluation Issue with Self-Converted DL3DV Chunks in MVSplat360

Hi @YaoXingbo, during testing, we only apply to scenes from the [DL3DV benchmark](https://huggingface.co/datasets/DL3DV/DL3DV-Benchmark). In this case, the testing data loader will only fetch those scenes that are recorded in [dl3dv_ctx_5v_tgt_56v.json](https://github.com/donydchen/mvsplat360/blob/main/assets/dl3dv_evaluation/dl3dv_ctx_5v_tgt_56v.json),...

About evaluation metric

Hi @zyqz97, when testing the PSNR and SSIM, we find it better to run all 56 frames through the SVD (modify the eval command by appending `model.refiner.svd_num_frames=56 model.refiner.test_time_attn_num_splits=4`), after that,...

Reference for `prepare_feat_proj_data_lists` and `warp_with_pose_depth_candidates` functions

Hi @emedinac, these two functions are designed to construct the cost volume. They are mainly based on and modified from [UniMatch](https://github.com/autonomousvision/unimatch). You can find a more formal description from the...

How to run mvsplat on custom data? and how to prepare dataset?

Hi @dev-moshpit, thanks for your interest in our work. For using custom data, you typically need to ensure that the camera intrinsic is normalized (detailed in https://github.com/donydchen/mvsplat/issues/28) and the camera...

Gaussian Param Prediction

Hi, @haowang020110, thanks for your interest in our work. This is similar to MVSplat, and you can find the same offset prediction in the MVSplat's [implementation](https://github.com/donydchen/mvsplat/blob/ef976a166da31e8392f2e5bee49bf66785e381e8/src/model/encoder/encoder_costvolume.py#L198). It is not something...

reference result is black

Hi @ChengLiDuoJi, this issue might be caused by the installation of `diff-gaussian-rasterization-modified`; try to install it manually by recursively cloning from the original GitHub project and then install it. More...