The evaluation results appear to be abnormal

[qq704554563]

Thank you for the wonderful work! However, after successfully running the evaluation code, the outputs shown below appear to be abnormal. Could you help me identify the problem? Additionally, could you provide the inference code that takes camera intrinsics and images as inputs and outputs the 3D Gaussians? This would allow me to visualize the scene freely with any rendering tools.

[botaoye]

Hi, could you please provide more details on how to run the code, such as the evaluate command? I will provide the script of saving 3D Gaussians soon.

[qq704554563]

Thank you for the reply. I ran the code as follows:
python -m src.main +experiment=re10k
mode=test wandb.name=re10k
dataset/[email protected]_sampler=evaluation
dataset.re10k.view_sampler.index_path=assets/evaluation_index_re10k_subset.json
checkpointing.load=./checkpoints/re10k.ckpt
test.save_image=true
evaluation_index_re10k_subset.json is the subset extracted from the evaluation_index_re10k.json
one sample is： "5aca87f95a9412c6": {"context": [58, 138], "target": [66, 104, 113], "overlap": 0.51419}

[botaoye]

Hi, the command seems to be correct. I double-checked the code and there is no problem. So there are several possible causes:

• Check if you have modified any part of the code, if so, maybe you can try to use the official code first and check if the result is correct.
• Check that the dataset is formatted correctly. pixelSplat provides test data that also works for us.