Open MultiPath opened 1 year ago
In the preprocessing code it links to the Google drive from the original SRN repo: https://github.com/NVlabs/eg3d/blob/main/dataset_preprocessing/shapenet_cars/run_me.py#L21 https://drive.google.com/uc?id=1bThUNtIHx4xEQyffVBSf82ABDDh2HlFn
However, it seems not the commonly used ShapeNet cars set used in other papers like PixelNeRF. This data contains around 2100 cars and each car has 250 cameras (also the intrinsics are incorrect).
The PixelNeRF shapenet, however, has 2400 cars in the training set and each car has 50 views, https://drive.google.com/drive/folders/1PsT3uKwqHHD2bEEHkIXB99AlIjtmrEiR which matches the text description in the paper
Can I confirm the shapenet model used in the paper is from which set?
I discovered the same issue, thanks for raising it. @MultiPath What makes you say that the intrinsics are incorrect?
In the preprocessing code it links to the Google drive from the original SRN repo: https://github.com/NVlabs/eg3d/blob/main/dataset_preprocessing/shapenet_cars/run_me.py#L21 https://drive.google.com/uc?id=1bThUNtIHx4xEQyffVBSf82ABDDh2HlFn
However, it seems not the commonly used ShapeNet cars set used in other papers like PixelNeRF. This data contains around 2100 cars and each car has 250 cameras (also the intrinsics are incorrect).
The PixelNeRF shapenet, however, has 2400 cars in the training set and each car has 50 views, https://drive.google.com/drive/folders/1PsT3uKwqHHD2bEEHkIXB99AlIjtmrEiR which matches the text description in the paper
Can I confirm the shapenet model used in the paper is from which set?