Creating a Navigable View for MVSplat Inference Results

[DataScientist-JPG]

I ran the following command:

python -m src.main +experiment=re10k checkpointing.load=checkpoints/re10k.ckpt mode=test dataset/view_sampler=evaluation test.compute_scores=true

The results have been generated in the outputs/test/re10k/ directory, which contains the following files:

0c4c5d5f751aabf5  2e109379f53bb221  656381bea665bf3d  a56ba2efb5e3fdd9  cd74b1244d112628  scores_all_avg.json
0d4de33c6888a754  322261824c4a3003  6771a51bf0cfce7f  a6fa92e8204f6118  e4f4574df7938f37  scores_lpips_all.json
1214f2a11a9fc1ed  34b0658a5c200cdf  67a69088a2695987  a9b3ff60b213e099  ed477bdf8582adff  scores_psnr_all.json
17d9303ee77c3a3d  41bcd011f99bfb66  84ab392d682f296b  aadc1e2dc74fd644  f7c0fa5b81552d35  scores_ssim_all.json
20d86cff490c0c42  57d25dafabb5a238  89ea49cd9865aeff  b4099665590548fc  f7d916b43193c181
21e794f71e31becb  57d3409bf04c4651  8adebbb68f2c3f84  bc95e5c7e357f1b7  fea544b472e9abd1
28e8300e004ab30b  5aca87f95a9412c6  9e2a8cc5f32dd46b  benchmark.json    ffa95c3b40609c76
29e0bfbad00f0d5e  6558c5f10d45a929  a47b88040452d7d9  c48f19e2ffa52523  peak_memory.json

I would like assistance in creating a navigable view for these output files. Suggestions for how to effectively visualize and navigate through the generated results would be greatly appreciated.

[donydchen]

Hi @DataScientist-JPG, this is mainly a research project and does not currently have an interactive demo. The file structure for the listed outputs can be found at

mvsplat/src/model/model_wrapper.py

Lines 202 to 211 in 31ac688

	(scene,) = batch["scene"]
	name = get_cfg()["wandb"]["name"]
	path = self.test_cfg.output_path / name
	images_prob = output.color[0]
	rgb_gt = batch["target"]["image"][0]
	# Save images.
	if self.test_cfg.save_image:
	for index, color in zip(batch["target"]["index"][0], images_prob):
	save_image(color, path / scene / f"color/{index:0>6}.png")

If you are more interested in the video results, you can find more details in the README. Cheers.