search

minghanqin / LangSplat

Official implementation of the paper "LangSplat: 3D Language Gaussian Splatting" [CVPR2024 Highlight]
https://langsplat.github.io/
Other
624 stars 64 forks source link

About the Ground Truth #28

Open Gloryseven opened 5 months ago

Gloryseven commented 5 months ago

I want to ask where the Ground Truth data in the paper comes from. I use the trained decoder to get the 'language_feature_dim3', and draw the array with shape of (H, W, 3) using the 'plt.imshow'. But the quality is worse than the GT image in the paper. Could you tell me how do you get and draw the GT image in the paper? Thank you!

minghanqin commented 5 months ago

I seek clarification regarding your reference to "GT"; are you referring to the visualization result of the feature map? It is imperative to acknowledge that if the feature map's output appears black, this phenomenon could be attributed to the feature map's value range, which extends from -1 to 1. Within this context, features ranging from -1 to 0 might be visualized as black. Try this visualization code.

img = (rendering_lf.permute(1,2,0).cpu().numpy() + 1) / 2

img = (img * 255).astype(np.uint8)

img = Image.fromarray(img)

img.save(os.path.join(render_path, '{0:05d}'.format(idx) + ".png"))