Downloads · 30 days
14
16% of all-time downloads
bambezius/sam3.1-soccernet-ball-streaming
sam3.1-soccernet-ball-streaming is a machine learning model from bambezius. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
Fine-tuned tracking components; requires the original gated facebook/sam3.1 checkpoint and its license/access agreement. Frozen image encoder weights are not duplicated here. These files are NOT standalone official pr…
Downloads · 30 days
14
16% of all-time downloads
All-time downloads
89
Public
Repo size
2.3 GB
Likes
0
Public
Click a slice to open those files.
.pt313 MB · 100%
From the Hugging Face model README
Fine-tuned tracking components; requires the original gated facebook/sam3.1
checkpoint and its license/access agreement. Frozen image encoder weights are
not duplicated here. These files are NOT standalone official predictor checkpoints.
One initial box per clip, carried spatial memory (initial + six recent frames), object pointers (initial + fifteen recent frames), 16-frame gradient windows. Missing annotations and initialization are excluded from loss/metrics. Targets are circular pseudo-masks derived from boxes, not manual segmentations.
best.pt: tracker selected by minimum validation loss.
last.pt: tracker, optimizer, RNG, epoch, and early-stopping state for resuming.
Patience is three non-improving validation epochs. See config/metrics for details.
Only load pickle checkpoints from trusted sources.
Use this repository's scripts/train_sam31_streaming.py --resume PATH/last.pt
to resume, or scripts/export_sam31_checkpoint.py --checkpoint PATH/best.pt
to merge with the pinned base for the official inference API.
Based on Meta SAM 3.1. Distributed under the included SAM License. Dataset media and annotations are not redistributed; dataset terms still apply.