Assembly Line Activity Recognition: Difference between revisions

Assembly Line Activity Recognition (view source)

2 bytes removed , 31 August 2022

m

no edit summary

20,290

edits

@@ Line 16: / Line 16: @@
 This project involves a complete machine learning activity recognition process: data capturing, labeling, developing the tools and training base, and finally parameter tuning and experimenting.
-The final trained model is capable of detecting activities from an assembly line video at 45 frames per second, with an accuracy, precision, recall and f1-score above 0.913 using transfer learning. The project was built around [https://pytorchvideo.org/ PyTorchVideo]'s SlowFast implementation; our custom software allows us to experiment with this network for our specific use case, modify parameters and test multiple training and tuning techniques in order to achieve the best results possible.
+The final trained model is capable of detecting activities from an assembly line video at 45 frames per second, with accuracy, precision, recall, and f1-score above 0.913 using transfer learning. The project was built around [https://pytorchvideo.org/ PyTorchVideo]'s SlowFast implementation; our custom software allows us to experiment with this network for our specific use case, modify parameters and test multiple training and tuning techniques in order to achieve the best results possible.
 The following video shows the inference results (in blue) as well as a manually labeled base (ground truth in green) for a test sample (these samples were not used for training).