Table 2 Comparisons of detection performance and inference speed for YOLOV4-MOD by using the proposed and baseline approach on model development test set data
From: Tile-based microscopic image processing for malaria screening using a deep learning approach
Model | Tile size | AP (%) | R (%) | Time (sec/img) | |
|---|---|---|---|---|---|
Train | Inference | ||||
YOLOV4-MOD@416 | 1088 \(\times\) 1088 | 1088 \(\times\) 1088 | 83.1 | 94.3 | 3.0 |
832 \(\times\) 832 | 85.5 | 95.1 | 4.0 | ||
608 \(\times\) 608 | 82.8 | 91.3 | 8.0 | ||
YOLOV4-MOD@512 | 1088 \(\times\) 1088 | 1088 \(\times\) 1088 | 81.8 | 90.5 | 4.0 |
832 \(\times\) 832 | 82.5 | 92.6 | 5.0 | ||
608 \(\times\) 608 | 78.2 | 91.7 | 9.0 | ||
YOLOV4-MOD@416 | 832 \(\times\) 832 | 1088 \(\times\) 1088 | 80.6 | 90.8 | 3.0 |
832 \(\times\) 832 | 85.0 | 93.5 | 4.0 | ||
608 \(\times\) 608 | 75.0 | 86.1 | 7.0 | ||
YOLOV4-MOD@512 | 832 \(\times\) 832 | 1088 \(\times\) 1088 | 66.1 | 82.3 | 4.0 |
832 \(\times\) 832 | 82.5 | 93.2 | 5.0 | ||
608 \(\times\) 608 | 79.6 | 92.7 | 10.0 | ||
YOLOV4-MOD@416 | 608 \(\times\) 608 | 1088 \(\times\) 1088 | 43.5 | 63.3 | 3.0 |
832 \(\times\) 832 | 66.7 | 82.5 | 3.5 | ||
608 \(\times\) 608 | 76.8 | 90.6 | 6.0 | ||
YOLOV4-MOD@512 | 608 \(\times\) 608 | 1088 \(\times\) 1088 | 55.3 | 85.7 | 3.0 |
832 \(\times\) 832 | 69.0 | 85.6 | 5.0 | ||
608 \(\times\) 608 | 77.9 | 90.1 | 8.5 | ||
Without tiling | |||||
YoloV4-MOD@512 | – | – | 79.78 | 80 | 0.25 |
YoloV4-MOD@416 | – | – | 72.8 | 76 | 0.2 |