Google Summer of Code: Week 2

The previous model trained was not performing well on the the testing dataset. So, I had to try a new approach for increasing the model accuracy as well as other parameters. When I tried training with other pre trained networks, after one epoch model started to overfit. So, I had to get creative.

I Used a  BiT model for training, only the last layer was trainable. The model was able to perform better than pervious pre-trained model based on MobileNet.

Regarding the VR Part. The WebXR setup was successful. I was able to set up a basic WebXR environment. TensorflowJs was used for model inference. I used react-framework for binding TensorflowJS and WebXR together. For access to onboard device sensors the WebXR content has to be served through HTTPS.

Model Training Results

MobileNet Based Model: loss: 0.9046 - acc: 0.7174 - f1_m: 0.2450 - precision_m: 301254.4277 - recall_m: 0.2275 - val_loss: 1.0795 - val_acc: 0.6811 - val_f1_m: 0.2366 - val_precision_m: 248592.8750 - val_recall_m: 0.2270

BiT Based Model:loss: 0.6331 - acc: 0.8733 - f1_m: 0.8899 - precision_m: 195757.4552 - recall_m: 0.9773 - val_loss: 1.2562 - val_acc: 0.7955 - val_f1_m: 0.9079 - val_precision_m: 126641.6953 - val_recall_m: 1.0619

Inference Time:

When the SavedModel is converted into TensorflowJS model using the converter few seconds get added up in the inference time. So far, the inference time is not that much and is close if not less than 1 sec.

That's all for day!!

Hope you Had great Week


Subscribe to Beyond the Blog

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
[email protected]