Pruned 3D CNNs with distillation improve worker activity recognition accuracy
PDNet: Exploring Pruning and Distillation for Real-Time Worker Activity Recognition in Industrial Assembly
Computer Vision and Pattern Recognition
Summary
Recognizing detailed human activities quickly and accurately is important for industrial assembly work. The authors present PDNet, a method that first shrinks a 3D convolutional neural network by removing less important parts, then teaches the smaller model using the original larger one to regain accuracy. This approach helps the smaller network perform much better than just pruning alone, achieving higher accuracy with a much smaller size and fewer computations. Their results suggest PDNet could be useful for recognizing worker actions in factories where computational resources are limited.
What this means in practice
- •For industrial automation engineers: Develop compact, accurate models for recognizing worker actions in real time on factory floors with limited computing power.
- •For embedded systems developers: Integrate pruned and distilled 3D CNN models into resource-constrained devices for fine-grained activity monitoring.
Authors
Malek Baba, Thinesh Thiyakesan Ponbagavathi, Alina Roitberg
Abstract
In this work, we address real-time fine-grained human activity recognition (HAR) in industrial assembly and introduce PDNet, a new framework that jointly exploits structured pruning and knowledge distillation for 3D convolutional networks. We start by training a 3D-CNN backbone in a supervised manner. Then, PDNet applies channel-wise structured pruning to obtain a more compact network. In the third step, our method is optimized using knowledge distillation from the large unpruned teacher to recover accuracy lost due to pruning. On the challenging MECCANO dataset with an X3D-M backbone, the pruned model without distillation reaches a Top-1 accuracy of 25.54 %, while PDNet raises this to 39.46 %, with only 6.90 MB model size and 3.78 GFLOPs. These results show that distillation within PDNet is highly effective in restoring pruned model performance and highlight its high potential for resource-constrained HAR in industrial assembly.