Table 3. Comparison of class-level accuracy with object detection methods (STD = standard deviation).
Method | Class-level accuracy | 5-fold CV STD | # of failures with no class |
---|---|---|---|
SSD | 33.98% | 0.1351 | 23 images |
Faster R-CNN with ResNet50 | 68.96% | 0.0923 | 9 images |
Our CNN model | 94.55% | 0.0104 | - |