Fig. 1. The basic blocks of ResNet and ResNeXt. (a) Basic block of ResNet; (b) Basic block of ResNeXt

Fig. 2. Workflow of the proposed algorithm

Fig. 3. Microscopic images of the three common pathogenic cells that cause vaginitis. (a) Mildew; (b) Trichomonas; (c) Clue cell

Fig. 4. The performance of the original DETR and the improved DETR on validation dataset

Fig. 5. The comparison of PR curves computed from the original model and the improved model. (a) PR curve of mAP; (b) PR curve of mildew; (c) PR curve of trichomonas; (d) PR curve of clue cell

Fig. 6. The detection results of the three common pathogenic cells. (a) Detection results of mildew; (b) Detection results of trichomonas; (c) Detection results of clue cell

Fig. 7. Comparison of detection results and attention weights visualization map. (a) Original image; (b) Ground truth; (c) Detection results of original DETR; (d) Attention weights visualization of the original DETR; (e) Attention weights visualization of the original DETR on the original image; (f) Detection results of the improved DETR; (g) Attention weights visualization of the improved DETR; (h) Attention weights visualization of the improved DETR on the original image

Table 1. The details of dataset split

Table 2. The comparison of mAP and AP results of the original model and the proposed model

Table 3. The comparison of precision and recall results of the original DETR and the proposed improved DETR