tarekziade/mobilenet-webnn

MobileNetV2 Image Classification using WebNN.

This model has been created with webnn-graph (https://github.com/tarekziade/webnn-graph), it's composed of 3 files:

• model.webnn : Graph definition
• model.weights : Model weights
• manifest.json: metadata, describes tensor shapes, data types and other things

To run this model, install PyWebNN and Pillow:

pip install pywebnn pillow onnxruntime

Warning : you need PyWebNN v0.5.0+

Grab the demo script here : https://huggingface.co/tarekziade/mobilenet-webnn/blob/main/demo.py

Then run the classifier on one of your images

➜  python demo.py test.jpg
======================================================================
MobileNetV2 Image Classification (Hugging Face Hub)
======================================================================
Image:   test.jpg
Model:   tarekziade/mobilenet-webnn
Backend: ONNX CPU

Downloading model from Hugging Face Hub...
   [CACHED] model.webnn
   [CACHED] model.weights
   [CACHED] manifest.json
   [OK] Downloaded (0.37ms)
   - Graph: model.webnn

Loading graph...
   [OK] Loaded (8.97ms)

Preprocessing image...
   [OK] (1, 3, 224, 224) (16.05ms)

Creating WebNN context...
   [OK] Context created (accelerated=False)

Running inference...
   [OK] Done (67.80ms)

Top 5 Predictions:
----------------------------------------------------------------------
   1. lesser panda                                        99.60%
   2. polecat                                              0.20%
   3. weasel                                               0.09%
   4. black-footed ferret                                  0.02%
   5. kit fox                                              0.01%

======================================================================
Performance Summary:
  - Model Download: 0.37ms
  - Graph Load:     8.97ms
  - Preprocessing:  16.05ms
  - Inference:      67.80ms
  - Total Time:     93.19ms
======================================================================

[OK] Done. Cache dir: /Users/tarekziade/.cache/webnn