Update Tensorflow MoViNets setup and examples to train streaming model on local dataset
Budget: £250 – £750 GBP
1) Update existing https://github.com/tensorflow/models/tree/master/official/projects/movinet training script/settings to train on locally stored video files using GPU (CUDA). Use https://www.crcv.ucf.edu/research/data-sets/ucf101/ dataset.
2) Script/command to export trained streaming model (trained in the previous step) as quantized TF Lite model
3) Script to test classification of random video from the dataset on TF Lite Streaming model (exported in the previous step) (can use https://github.com/tensorflow/examples/blob/master/lite/examples/video_classification/raspberry_pi/ as the starting point)
4) Update https://github.com/tensorflow/models/blob/master/official/projects/movinet/movinet_tutorial.ipynb "Fine-Tune a Base Model with the TensorFlow Model Garden" example to fine-tune streaming model using locally stored videos (use same ucf101 dataset)
2) Script/command to export trained streaming model (trained in the previous step) as quantized TF Lite model
3) Script to test classification of random video from the dataset on TF Lite Streaming model (exported in the previous step) (can use https://github.com/tensorflow/examples/blob/master/lite/examples/video_classification/raspberry_pi/ as the starting point)
4) Update https://github.com/tensorflow/models/blob/master/official/projects/movinet/movinet_tutorial.ipynb "Fine-Tune a Base Model with the TensorFlow Model Garden" example to fine-tune streaming model using locally stored videos (use same ucf101 dataset)