search

Guillem96 / efficientdet-tf

Custom implementation of EfficientDet https://arxiv.org/abs/1911.09070
GNU General Public License v3.0
96 stars 23 forks source link

Will TFLite export be supported? #15

Closed lessw2020 closed 3 years ago

lessw2020 commented 4 years ago

I wasn't able to save a model to test as there was an error with the bounding box code (shown below) but my main question was to find out if TFLite export and quantization will be supported for this project? (need to run on mobile).

Thanks very much.
Related - but here is the error when I tried to save the model after running the example notebook: `-- TypeError Traceback (most recent call last)

in () ----> 1 model.save('tester',include_optimizer=False) 26 frames /usr/local/lib/python3.6/dist-packages/efficientdet/utils/bndbox.py in nms(boxes, class_scores, score_threshold) 182 all_scores = [] 183 --> 184 for batch_idx in range(batch_size): 185 body_fn = partial(body, batch_idx=batch_idx) 186 # For each class, get the effective boxes, labels and scores TypeError: 'Tensor' object cannot be interpreted as an integer`
Guillem96 commented 4 years ago

Save the weights only.

Also, the export to saved-model is incoming.

veb-101 commented 4 years ago

Any updates on saved_model export?