Open Access
Table 2
Best single and multi-class models comparing original and tiny YOLOv4.
| Model | Network | Best models | mAP (Validation) (%) | mAP (Test) (%) |
mAP (Average) (%) |
Prediction time (ms) | Training time (hrs) | Memory size (MB) |
|---|---|---|---|---|---|---|---|---|
| Single class (microrack) | Original | Model 2 | 100.0 | 98.8 | 99.9 | 62.9 | 26.0 | 244 |
| Tiny | Model 7 | 99.3 | 99.6 | 99.4 | 26.9 | 2.9 | 22.4 | |
| Single class (dark region) | Original | Model 1 | 99.4 | 88.5 | 93.9 | 41.8 | 14.7 | 244 |
| Tiny | Model 7 | 93.7 | 71.4 | 82.6 | 25.8 | 3.0 | 22.4 | |
| Multi-class (Microcrack and dark region) | Original | Model 2 | 99.5 | 98.0 | 98.8 | 62.9 | 26.3 | 244 |
| Tiny | Model 14 | 94.6 | 85.7 | 90.1 | 27.8 | 5.0 | 23.3 |
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.