EdgeTAM (ONNX), as the Q-P1NG app downloads it

An unchanged copy of the full-precision ONNX export of EdgeTAM (Meta, "EdgeTAM: On-Device Track Anything Model", CVPR 2025; Apache-2.0) made by onnx-community: onnx-community/EdgeTAM-ONNX at commit 9c77c7bff7fd0f3079585fa17af7f730ddc531ed, from the transformers port yonigozlan/EdgeTAM-hf. LICENSE is the Apache License 2.0 of facebookresearch/EdgeTAM at commit 7711e012a30a2402c4eaab637bdb00a521302c91. It is kept here so that the Q-P1NG app's selection pack downloads from a place that does not change; the app checks every file's SHA-256 before using it. Nothing was modified.

File	SHA-256	Bytes
onnx/vision_encoder.onnx	ed068218eba96760fe02d04ce899c449660ac813a088d80ed7f42c8bb01e7cec	192225
onnx/vision_encoder.onnx_data	21e75dba7077dfcb53e8c9a6e99977156f2240ff3f1f9cddc43d66aa1ecb528e	19532576
onnx/prompt_encoder_mask_decoder.onnx	d3668299ec3edf70fbb139ec642b54bf3d4be453fd1b688a6b5938e0856fe546	213114
onnx/prompt_encoder_mask_decoder.onnx_data	dfa2125e30d08d388732f20c18fb63ab0f0f590cd270eb307f0c2b65919d1be8	20958208
preprocessor_config.json	126d931a4698900d085aae8c2cf57ecc6923fa60b0bf425e490189f14c7200ee	720
LICENSE	c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4	11357

Input and output

• onnx/vision_encoder.onnx: pixel_values float32 [1, 3, 1024, 1024], RGB resized to 1024 x 1024, rescaled by 1/255 and normalised with ImageNet's mean [0.485, 0.456, 0.406] and deviation [0.229, 0.224, 0.225]; outputs image_embeddings.0 [1, 32, 256, 256], image_embeddings.1 [1, 64, 128, 128], image_embeddings.2 [1, 256, 64, 64].
• onnx/prompt_encoder_mask_decoder.onnx: input_points float32 [1, 1, N, 2] (pixels of the 1024 x 1024 picture), input_labels int64 [1, 1, N] (1 on the object, 0 off it, -10 padding), input_boxes float32 [1, B, 4] (B = 0 or 1) and the three embeddings; outputs iou_scores [1, 1, 3], pred_masks [1, 1, 3, 256, 256] (logits) and object_score_logits [1, 1, 1] (above 0: an object is there).

License

Apache License 2.0: see LICENSE. EdgeTAM is by Meta Platforms, Inc. and affiliates.

Citation

@inproceedings{zhou2025edgetam,
  title={EdgeTAM: On-Device Track Anything Model},
  author={Zhou, Chong and Zhu, Chenchen and Xiong, Yunyang and Suri, Saksham and Xiao, Fanyi and Wu, Lemeng and
          Krishnamoorthi, Raghuraman and Dai, Bo and Loy, Chen Change and Chandra, Vikas and Soran, Bilge},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2025}
}