Models
From catalog/models.json. The fit column says where each model is expected to run: device:ram>=512k needs that much board RAM, +int8 needs model.quantize: int8, device:psram needs a board with PSRAM, edge and cloud run through westwind serve. The memory plan is the planned activation memory and the size of the ExecuTorch program for each conversion that has been measured.
| Model id | Task | Source | Size | Input | Fit | Memory plan | License |
|---|---|---|---|---|---|---|---|
mediapipe/face_detector/blaze_face_short_range | detection | mediapipe | 224 KiB | [1, 128, 128, 3] NHWC float32 | device:psramedgecloud | fp32: 1.7 MiB, .pte 487 KiB int8-cortex-m: 1.0 MiB, .pte 454 KiB | Apache-2.0 |
mediapipe/image_segmenter/selfie_segmenter | segmentation | mediapipe | 244 KiB | [1, 256, 256, 3] NHWC float32 | device:psramedgecloud | fp32: 4.3 MiB, .pte 550 KiB | Apache-2.0 |
mediapipe/image_segmenter/deeplab_v3 | segmentation | mediapipe | 2.7 MiB | see model | edgecloud | Apache-2.0 | |
mediapipe/pose_landmarker/pose_landmarker_lite | pose | mediapipe | 5.5 MiB | see model | edgecloud | The landmark model converts; the detector uses sparse weights the converter does not read yet. Run it with westwind serve (LiteRT). | Apache-2.0 |
mediapipe/hand_landmarker/hand_landmarker | pose | mediapipe | 7.5 MiB | see model | edgecloud | Apache-2.0 | |
mediapipe/face_landmarker/face_landmarker | pose | mediapipe | 3.6 MiB | see model | edgecloud | Apache-2.0 | |
mediapipe/audio_classifier/yamnet | classification | mediapipe | 3.9 MiB | see model | device:psramedgecloud | Apache-2.0 | |
mediapipe/object_detector/efficientdet_lite0 | detection | mediapipe | 4.4 MiB | see model | edgecloud | Apache-2.0 | |
builtin/tiny_digit_net | classification | builtin | [1, 1, 28, 28] NCHW float32 | device:ram>=128k | Apache-2.0 | ||
builtin/anomaly_net | anomaly | builtin | [1, 32] NC float32 | device:ram>=32k | Apache-2.0 | ||
builtin/kws_net | keyword-spotting | builtin | [1, 1, 49, 10] NCHW float32 | device:ram>=128k | Apache-2.0 | ||
hf-space/eoinedge/sam3 | segmentation | huggingface | see model | cloud | see Space |