Argus-Lite

Argus-Lite is Argus on a frozen EUPE-ViT-S backbone (Zhu et al., Meta FAIR, arXiv:2603.22387): 38M parameters in total, of which 16.5M are the heads. It shares Argus's code and API; config.json selects the backbone width and the head variants.

Task	Head	Output	Trained on	Result
Classification	Linear softmax on the L2-normalized CLS token	1,000 ImageNet classes	ImageNet-1k	79.13% top-1, 95.53% top-5 (val)
Segmentation	BatchNorm and 1×1 conv	150 ADE20K classes	ADE20K	41.9 mIoU (val)
Depth	DPT decoder on blocks 2, 5, 8 and 11, 256 log-spaced bins over 0.001–10 m	Metric depth	NYU Depth V2	RMSE 0.321, abs rel 0.103, δ<1.25 0.898 (654-image Eigen test split, Eigen crop, evaluate.py)
Detection	Cofiber pyramid with a text-aligned cosine classifier, FCOS decoding	80 COCO classes	COCO 2017	27.3 mAP, 49.6 AP@50 (val2017)
Correspondence	None: cosine matching of patch features	Patch matches	—	—

Class-agnostic AR@100 across 20 RF100-VL domains is 0.266, with the per-domain results in rf100vl_results.json; cls_val_eval.json and coco_val_eval.json hold the classification and detection breakdowns.

Usage

from PIL import Image
from transformers import AutoModel

model = AutoModel.from_pretrained("phanerozoic/argus-lite", trust_remote_code=True)
image = Image.open("image.jpg").convert("RGB")

model.classify(image, top_k=5)          # [{"class_id", "class_name", "score"}, ...]
model.segment(image)                    # [H, W] ADE20K class indices
model.depth(image)                      # [H, W] depth in meters
model.detect(image, score_thresh=0.3)   # [{"box", "score", "label", "class_name"}, ...]
model.embed(image)                      # [384] L2-normalized CLS embedding
model.perceive(image)                   # classification, segmentation and depth
model.correspond(image, other_image)    # {"matches", "scores", "grid"}

Each method also takes a list of images and returns a list of results. Images are resized to the resolution of the head: 224 for classification and embedding, 512 for segmentation, 416 for depth and 768 for detection. Segmentation and depth maps come back at each image's own height and width, and boxes in its pixel coordinates. segment() and depth() return per-pixel confidence and depth standard deviation with return_confidence=True. The depth head is trained on indoor scenes and predicts at most 10 m. quantize_int8(), compile(), export_onnx() and evaluate.py work as in Argus.

License

The EUPE-ViT-S weights are released under the FAIR Noncommercial Research License. The checkpoint contains them and is released under the same license.

Citation

@misc{zhu2026eupe,
  title={Efficient Universal Perception Encoder},
  author={Zhu, Chenchen and Suri, Saksham and Jose, Cijo and Oquab, Maxime and Szafraniec, Marc and Wen, Wei and Xiong, Yunyang and Labatut, Patrick and Bojanowski, Piotr and Krishnamoorthi, Raghuraman and Chandra, Vikas},
  year={2026},
  eprint={2603.22387},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}