L2CS-Net
(Ahmednull) gaze estimation, converted to
LiteRT
and
running
fully on the
CompiledModel
GPU
(ML Drift) on Android. Predicts where a centered face is looking
(yaw/pitch). ResNet50 backbone trained on Gaze360.
On-device (Pixel 8a, Tensor G3 — verified)
nodes on GPU
139 / 139
LITERT_CL (full residency)
inference
~3 ms
(448×448)
size
47.9 MB (fp16)
accuracy
device-vs-PyTorch corr
0.9999
, gaze angle within ~0.1°
face[1,3,448,448] (ImageNet-normalized) →[GPU: ResNet50]→ yaw[1,90], pitch[1,90] (softmax over angle bins)
Pure CNN (ResNet50 + 2 FC heads). Two numerically-exact ResNet fixes:
stem
MaxPool2d(3,s2,p1)
→ zero-pad + valid max-pool
— PyTorch's max-pool pads with
-inf
→ a
PADV2
the Mali delegate won't delegate (compile fail); since the pool follows a ReLU, a 0-pad is exactly
equivalent →
PAD
, full GPU residency.
L2CS-Gaze360-LiteRT huggingface.co is an AI model on huggingface.co that provides L2CS-Gaze360-LiteRT's model effect (), which can be used instantly with this litert-community L2CS-Gaze360-LiteRT model. huggingface.co supports a free trial of the L2CS-Gaze360-LiteRT model, and also provides paid use of the L2CS-Gaze360-LiteRT. Support call L2CS-Gaze360-LiteRT model through api, including Node.js, Python, http.
L2CS-Gaze360-LiteRT huggingface.co is an online trial and call api platform, which integrates L2CS-Gaze360-LiteRT's modeling effects, including api services, and provides a free online trial of L2CS-Gaze360-LiteRT, you can try L2CS-Gaze360-LiteRT online for free by clicking the link below.
litert-community L2CS-Gaze360-LiteRT online free url in huggingface.co:
L2CS-Gaze360-LiteRT is an open source model from GitHub that offers a free installation service, and any user can find L2CS-Gaze360-LiteRT on GitHub to install. At the same time, huggingface.co provides the effect of L2CS-Gaze360-LiteRT install, users can directly use L2CS-Gaze360-LiteRT installed effect in huggingface.co for debugging and trial. It also supports api for free installation.
L2CS-Gaze360-LiteRT install url in huggingface.co: