L2CS-Net (Gaze Estimation)

Licensing scope: See the license and provenance notice before relying on this card's license metadata for pretrained-weight redistribution or commercial use. Existing valid grants are preserved.

Model Description

L2CS-Net regresses gaze direction (pitch, yaw) from a face crop. It formulates gaze estimation as a 90-bin classification problem over each axis (4°/bin resolution covering [-180°, +180°]), then computes the expected value across bins for a continuous angle output. ResNet50 backbone, two parallel FC heads.

Reported accuracy (from the original L2CS-Net paper):

• Gaze360 test split: ~3.92° MAE
• MPIIFaceGaze leave-one-out: ~4.16° MAE

These are state-of-the-art numbers for gaze-from-face-crop estimation (2022); for context, geometric iris-eye approaches typically land at 8-15° MAE on the same benchmarks.

Model Details

• Architecture: ResNet50 + dual classification heads (2 × 90-bin)
• Input: 224 × 224 RGB face crop, ImageNet-normalized
• Output: pitch, yaw in radians (head-centric frame)
• Bin resolution: 4° per bin, covering [-180°, +180°]
• Backbones available: ResNet50 (default), ResNet18 (lighter)
• Framework: PyTorch (port of upstream MIT code)

Training data (upstream)

• Gaze360 (Kellnhofer et al., 2019): in-the-wild gaze annotations with 360° head pose coverage. ~127k images.
• MPIIFaceGaze (Zhang et al., 2017): unconstrained office captures with screen-targeted gaze ground truth. ~213k images.

The upstream maintainer trains separate checkpoints for each dataset. The checkpoint shipped by Py-Feat is Gaze360-trained: l2cs_gaze360_resnet50.safetensors, converted from L2CSNet_gaze360.pkl. MPIIFaceGaze is a separate upstream training and evaluation configuration, not a second training corpus for this checkpoint.

Model Sources

• Original repository (MIT): Ahmednull/L2CS-Net
• Paper: arXiv:2203.03339
• Pretrained weights (upstream): Google Drive folder linked from the upstream README (L2CSNet_gaze360.pkl). Py-Feat hosts a re-packaged .safetensors version on this repo to avoid the pickle (.pkl) deserialization path on user machines. The conversion is documented at scripts/convert_l2cs_pickle_to_safetensors.py in the py-feat repo; no architecture or weight values are modified.

Acknowledgements

This distribution is a re-host of the official L2CS-Net weights trained by Abdelrahman, Hempel, Khalifa, and Al-Hamadi (Otto-von-Guericke University Magdeburg). All training, hyperparameter selection, and benchmark numbers are credited to the original authors. The py-feat project provides only:

• A PyTorch port of the inference code at feat/gaze_detectors/l2cs/l2cs_model.py
• This re-packaged .safetensors artifact for downstream safety
• Integration with feat.Detector and feat.MPDetector's pipelines

The shipped checkpoint's training data are credited to:

• Gaze360 — Kellnhofer, Recasens, Stent, Matusik, Torralba (MIT)

The separate MPIIFaceGaze configuration uses data from Zhang, Sugano, Fritz, and Bulling (MPI Saarbrücken).

Citation

@article{l2csnet2022,
  title={L2CS-Net: Fine-Grained Gaze Estimation in Unconstrained Environments},
  author={Abdelrahman, Ahmed and Hempel, Thorsten and Khalifa, Aly and
          Al-Hamadi, Ayoub},
  journal={arXiv preprint arXiv:2203.03339},
  year={2022}
}

License and provenance

The original implementation is MIT-licensed. The historical model-card designation was MIT. The current other label links this scope notice to distinguish MIT code from the separate Gaze360 research terms; it is not a new combined license or a revocation of existing valid grants. Conversion to safetensors does not create additional rights.

The shipped checkpoint's Gaze360 source license, section 2(3) expressly excludes commercial applications of models trained on the dataset. Its restrictions on use and distribution require separate assessment of public weight sharing. The MIT implementation license does not waive these express conditions, and this clarification does not revoke existing valid grants.

See the v1 component notice and dataset register for checkpoint provenance and the outstanding permission review. Terms for the separate MPIIFaceGaze models must be assessed for those checkpoints.

This notice clarifies scope; it does not revoke existing valid grants or create a new license for third-party material. The Gaze360 restriction expressly addresses trained models; it is not inferred from automatic license inheritance. A software license or model-card badge alone does not establish all checkpoint redistribution or commercial-use permissions.

The applicable acquisition terms and any separate permissions must be checked for the exact checkpoint; today's source notice does not establish the history of every upstream grant.