search

lkeab / gaussian-grouping

[ECCV'2024] Gaussian Grouping for open-world Anything reconstruction, segmentation and editing.
https://arxiv.org/abs/2312.00732
Apache License 2.0
681 stars 42 forks source link

TypeError converting np.ndarray of type numpy.object_ #15

Open kleinfour opened 10 months ago

kleinfour commented 10 months ago

Trying to train.py on my colmap dataset but run into this error: File "C:\Users\User\gaussian-grouping\utils\camera_utils.py", line 54, in loadCam objects=torch.from_numpy(np.array(cam_info.objects))) TypeError: can't convert np.ndarray of type numpy.object_. The only supported types are: float64, float32, float16, complex64, complex128, int64, int32, int16, int8, uint8, and bool.

print(cam_info.objects) prints None [10/01 18:25:21] I trained my own segmentation masks and they are located in the 'object_mask' directory with the rest of my scene's data.

Maybe because I installed using pytorch 2.1.2 and compiled with cuda toolkit 12.1? I'm running Windows 11 and there were far too many issues using the cuda 11.3 build. I appreciate any help with debugging this.

ymq2017 commented 10 months ago

Hi, it looks like you do not load the correct object mask path. You can add a breakpoint here and check os.path.exists(object_path).

kleinfour commented 10 months ago

Hi, it looks like you do not load the correct object mask path. You can add a breakpoint here and check os.path.exists(object_path).

Hi, thanks for the reply. I found that my filenames, which are in the form "1305031452.791720.png" was causing an issue with the file due to the "." in the middle of the name. I fixed it by changing this line to use rsplit instead: image_name = os.path.basename(image_path).rsplit(".", 1)[0]