YOLOv7
YOLOv7
Embedded Files
YOLOv7 is a powerful object detection algorithm that can run on images, video, and live camera feed. The backbone of the model uses an Extended Efficient Layer Aggregation Network (E-ELAN), which improves the network's ability to learn continuously while the maintaining the gradient's original path. Predicted outputs are contained in the model heads, with the final output stored in the lead head. You can find the paper here or read it below.
Methods
Methods
Model: We used YOLOv7-tiny instead of YOLOv7 because it is built for edge devices, is less resource-intensive, and has a faster computing time.
Data source: We manually collected videos of Spencer performing the procedure with varying skill levels. We then labeled them with CVAT.ai. We sourced additional hand data from existing datasets.
Data Augmentation: We experimented with several image augmentation techniques (Mixup, Cutout + Mixup, and CutMix). Compared to the other image augmentation methods, CutMix augmentation saw a visible improvement in obstruction problems.
Training Methods: Transfer learning uses a small learning rate on top of a pre-trained model. This method was most effective.
Page updated
Google Sites
Report abuse