search

ShijieZhou-UCLA / feature-3dgs

[CVPR 2024 Highlight] Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields
Other
381 stars 25 forks source link

really low resolution for gt_feature_map and feature map SAM encoding #47

Closed wrencanfly closed 3 weeks ago

wrencanfly commented 3 weeks ago

I followed the instructions and tried using the teatime dataset.

what I did:

using SAM: python export_image_embeddings.py --checkpoint checkpoints/sam_vit_h_4b8939.pth --model-type vit_h --input ../../data/DATASET_NAME/images --output ../../data/OUTPUT_NAME/sam_embeddings

train: python train.py -s data/DATASET_NAME -m output/OUTPUT_NAME -f sam --speedup --iterations 7000

render: python render.py -s data/DATASET_NAME -m output/OUTPUT_NAME --iteration 7000

image

ShijieZhou-UCLA commented 3 weeks ago

Since the resolution for SAM's GT feature map is very low, you can comment out this line to get the same resolution feature map as your rendered image size: https://github.com/ShijieZhou-UCLA/feature-3dgs/blob/6c570d6e7e4375129e57a588a206e6f629be2ff1/render.py#L137