Few-shot object detector to TFLite

ml-gde / e2e-tflite-tutorials

Project tracking of the "Mobile ML Working Group", for the End-to-End TensorFlow Lite tutorials.

Apache License 2.0

133 stars 26 forks source link

Few-shot object detector to TFLite #24

Closed sayakpaul closed 4 years ago

sayakpaul commented 4 years ago

@khanhlvg the TFOD API few-shot training example uses this checkpoint ssd_resnet50_v1_fpn_640x640_coco17_tpu-8.

Would be nice to see if that fine-tuned model is convertible to TFLite using the recommended instructions. WDYT?

khanhlvg commented 4 years ago

Sure, we can try with that model.

sayakpaul commented 4 years ago

@khanhlvg the only issue I found there is a lack of guidance to serialize a model trained using the eager mode. I have filed an issue for that. I will keep this thread updated once I have something.

khanhlvg commented 4 years ago

I see. TFLite team also have a plan to work on a few shot training e2e notebook including the conversion to TFLite. Therefore you can prioritize working on the MobileDet models as it's unlikely that anyone will work on it soon.

sayakpaul commented 4 years ago

Closing the issue for now then. Feel free to open it up.