Repository Issue Activity (beta)

#604 SAM3 mask exemplars: claimed in the paper, unusable in the release

#507 RuntimeError: mat1 and mat2 must have the same dtype, but got BFloat16 and Float during image inference

#597 How I can run this on mac No module named 'triton'

#590 Constant "We're at capacity right now" error on Meta AI Demos playground

#595 How were the choices of memory_temporal_stride_for_eval=1 and num_maskmem=7 made?

#163 How to fine-tune on my own image/video datasets?

#594 In the same scenario, most of the reasoning can be correct, while a small portion yields no result.

#596 Is CUDA 12.8 a hard requirement, or just the dev environment default?

#495 Tracking long videos

#577 ValueError when running sam3_image_interactive.ipynb

#591 SAM3 Create 3D Bodies has a bug

#587 An error occurs in CPU-only environment due to hardcoded 'cuda' device

#586 Cool trick.

#585 [SA-Co eval Bug?] Custom CGF1 evaluator penalizes predictions overlapping with crowd annotations (iscrowd=1)

#180 ImportError: cannot import name 'Sam3Processor' from 'transformers'

#584 Any ideas on how to segment moving vehicles as separate class?

#336 How to predict all segmentation results in an image?

#583 image to 3d not convert please fix this issue

#582 Potential workaround for SAM3.1 checkpoint downloads hanging indefinitely

#528 Abnormal floating masks when using SAM 3.1 for MOT, leading to oversized bounding boxes

#581 SAM3 in Deepstream

#580 anyone successfully fine-tuned sam3 on a custom dataset?

#579 Would SA-CO/VIDEO dataset be released?

#578 usage example for segmenting lidar

#575 Reduced Triton sigmoid focal loss returns NaN gradients for gamma=0 when logits saturate