Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.
  To pick up a draggable item, press the space bar. While dragging, use the arrow keys to move the item. Press space again to drop the item in its new position, or press escape to cancel.

Question

Hi!
I am trying to evaluate my trained YOLO model on a (custom) test dataset. Yet it seems that I get very similar results for different IOU-thresholds e.g. 0.3 and 0.8. Also for threshold values of 0.1 and 0.2 precision and recall are 1. These results don't seem right.
Also, could anyone explain what influence the threshold has on the mAP value, i.e. if the threshold is larger than 0.5 shouldn't the mAP values decrease?

Thank you in advance for your help!

Below are the executed commands and outputs:

!python val.py --weights ../../content/drive/MyDrive/Masterarbeit/Class_Alldata_LISA1280_Bosch1280_DTLD1280/weights/best.pt --data coco128.yaml --iou 0.2 --img 1280 --verbose --save-hybrid

val: data=/content/yolov5/data/coco128.yaml, weights=['../../content/drive/MyDrive/Masterarbeit/Class_Alldata_LISA1280_Bosch1280_DTLD1280/weights/best.pt'], batch_size=32, imgsz=1280, conf_thres=0.001, iou_thres=0.2, task=val, device=, single_cls=False, augment=False, verbose=True, save_txt=True, save_hybrid=True, save_conf=False, save_json=False, project=runs/val, name=exp, exist_ok=False, half=False
YOLOv5 🚀 v6.0-55-g0155548 torch 1.9.0+cu111 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)

Fusing layers...
Model Summary: 392 layers, 47047521 parameters, 0 gradients, 115.5 GFLOPs
val: Scanning '../../content/drive/MyDrive/test_images_combined/labels.cache' images and labels... 14623 found, 0 missing, 15 empty, 15 corrupted: 100% 14625/14625 [00:00<?, ?it/s]

           Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100% 457/457 [08:27<00:00,  1.11s/it]
             all      14610      38674          1          1      0.995      0.995
             Red      14610      14463          1          1      0.995      0.995
          Yellow      14610       1437          1          1      0.995      0.995
           Green      14610      20472          1          1      0.995      0.995
             off      14610       2302          1          1      0.995      0.995

Speed: 0.3ms pre-process, 25.1ms inference, 1.2ms NMS per image at shape (32, 3, 1280, 1280)
Results saved to runs/val/exp4
14604 labels saved to runs/val/exp4/labels

!python val.py --weights ../../content/drive/MyDrive/Masterarbeit/Class_Alldata_LISA1280_Bosch1280_DTLD1280/weights/best.pt --data coco128.yaml --iou 0.3 --img 1280 --verbose --save-txt

val: data=/content/yolov5/data/coco128.yaml, weights=['../../content/drive/MyDrive/Masterarbeit/Class_Alldata_LISA1280_Bosch1280_DTLD1280/weights/best.pt'], batch_size=32, imgsz=1280, conf_thres=0.001, iou_thres=0.3, task=val, device=, single_cls=False, augment=False, verbose=True, save_txt=True, save_hybrid=False, save_conf=False, save_json=False, project=runs/val, name=exp, exist_ok=False, half=False
YOLOv5 🚀 v6.0-55-g0155548 torch 1.9.0+cu111 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)

Fusing layers...
Model Summary: 392 layers, 47047521 parameters, 0 gradients, 115.5 GFLOPs
val: Scanning '../../content/drive/MyDrive/test_images_combined/labels.cache' images and labels... 14623 found, 0 missing, 15 empty, 15 corrupted: 100% 14625/14625 [00:00<?, ?it/s]

           Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100% 457/457 [08:23<00:00,  1.10s/it]
             all      14610      38674      0.618      0.418       0.42      0.232
             Red      14610      14463      0.681      0.502      0.522      0.271
          Yellow      14610       1437      0.411      0.384      0.337      0.191
           Green      14610      20472       0.71      0.636      0.627      0.341
             off      14610       2302       0.67      0.149      0.195      0.125

Speed: 0.4ms pre-process, 25.1ms inference, 0.9ms NMS per image at shape (32, 3, 1280, 1280)
Results saved to runs/val/exp5
14503 labels saved to runs/val/exp5/labels

!python val.py --weights ../../content/drive/MyDrive/Masterarbeit/Class_Alldata_LISA1280_Bosch1280_DTLD1280/weights/best.pt --data coco128.yaml --iou 0.8 --img 1280 --verbose --save-txt

val: data=/content/yolov5/data/coco128.yaml, weights=['../../content/drive/MyDrive/Masterarbeit/Class_Alldata_LISA1280_Bosch1280_DTLD1280/weights/best.pt'], batch_size=32, imgsz=1280, conf_thres=0.001, iou_thres=0.8, task=val, device=, single_cls=False, augment=False, verbose=True, save_txt=True, save_hybrid=False, save_conf=False, save_json=False, project=runs/val, name=exp, exist_ok=False, half=False
YOLOv5 🚀 v6.0-55-g0155548 torch 1.9.0+cu111 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)

Fusing layers...
Model Summary: 392 layers, 47047521 parameters, 0 gradients, 115.5 GFLOPs
val: Scanning '../../content/drive/MyDrive/test_images_combined/labels.cache' images and labels... 14623 found, 0 missing, 15 empty, 15 corrupted: 100% 14625/14625 [00:00<?, ?it/s]

           Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100% 457/457 [08:49<00:00,  1.16s/it]
             all      14610      38674      0.612      0.418      0.408      0.221
             Red      14610      14463      0.674      0.502      0.502      0.255
          Yellow      14610       1437      0.405      0.384      0.335      0.188
           Green      14610      20472      0.704      0.636      0.612      0.322
             off      14610       2302      0.664      0.149      0.184      0.119

Speed: 0.4ms pre-process, 25.2ms inference, 0.9ms NMS per image at shape (32, 3, 1280, 1280)
Results saved to runs/val/exp7
14503 labels saved to runs/val/exp7/labels

Additional

No response