b站视频:https://www.bilibili.com/video/BV1fD421j7hQ/
【腾讯文档】RT-DETR改进前后数据
目录
1.RT-DETR-r50
1.1.last
rtdetr-r50 summary: 480 layers, 41964383 parameters, 0 gradients, 129.6 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 380/380 [00:17<00:00, 21.27it/s]
all 3036 3427 0.699 0.685 0.659 0.507
weigh 3036 337 0.68 0.721 0.692 0.537
height measure 3036 802 0.681 0.5 0.527 0.356
drop ball 3036 820 0.601 0.741 0.609 0.475
size measure 3036 601 0.672 0.631 0.619 0.481
record 3036 867 0.863 0.834 0.847 0.688
Speed: 0.1ms preprocess, 4.3ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp
1.2.best
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all2/labels/val.cache... 3036 images, 0 backgrounds, 0 corrupt: 100%|██████████| 3036/3036 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 759/759 [00:26<00:00, 28.76it/s]
all 3036 3427 0.71 0.681 0.658 0.507
weigh 3036 337 0.688 0.721 0.691 0.535
height measure 3036 802 0.693 0.481 0.525 0.356
drop ball 3036 820 0.61 0.733 0.606 0.473
size measure 3036 601 0.687 0.633 0.622 0.482
record 3036 867 0.87 0.835 0.847 0.69
Speed: 0.1ms preprocess, 6.8ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/val/exp
2.RT-DETR-r50-filter
2.1.last
rtdetr-r50 summary: 480 layers, 41964383 parameters, 0 gradients, 129.6 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 365/365 [00:17<00:00, 21.07it/s]
all 2915 3275 0.721 0.654 0.663 0.511
weigh 2915 337 0.678 0.668 0.657 0.512
height measure 2915 726 0.669 0.482 0.536 0.385
drop ball 2915 757 0.642 0.665 0.614 0.463
size measure 2915 589 0.761 0.643 0.671 0.519
record 2915 866 0.856 0.813 0.838 0.677
Speed: 0.1ms preprocess, 4.3ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp3
2.2.best
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all3_filter/labels/val.cache... 2915 images, 0 backgrounds, 0 corrupt: 100%|██████████| 2915/2915 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 729/729 [00:25<00:00, 28.35it/s]
all 2915 3275 0.722 0.654 0.662 0.511
weigh 2915 337 0.683 0.671 0.654 0.511
height measure 2915 726 0.661 0.482 0.54 0.388
drop ball 2915 757 0.643 0.666 0.613 0.464
size measure 2915 589 0.764 0.635 0.665 0.515
record 2915 866 0.858 0.816 0.839 0.677
Speed: 0.1ms preprocess, 6.8ms inference, 0.0ms loss, 0.2ms postprocess per image
3.RT-DETR-r50-filter-data-enhance
3.1.last
rtdetr-r50 summary: 480 layers, 41964383 parameters, 0 gradients, 129.6 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 407/407 [00:18<00:00, 21.50it/s]
all 3252 3624 0.741 0.687 0.667 0.53
weigh 3252 674 0.876 0.653 0.669 0.554
height measure 3252 726 0.642 0.519 0.518 0.376
drop ball 3252 758 0.62 0.781 0.673 0.52
size measure 3252 589 0.781 0.628 0.629 0.489
record 3252 877 0.787 0.855 0.848 0.713
Speed: 0.1ms preprocess, 4.3ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp5
3.2.best
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all4_data_enhance/labels/val.cache... 3252 images, 0 backgrounds, 0 corrupt: 100%|██████████| 3252/
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 813/813 [00:28<00:00, 29.03it/s]
all 3252 3624 0.739 0.684 0.663 0.527
weigh 3252 674 0.876 0.651 0.671 0.556
height measure 3252 726 0.641 0.514 0.512 0.37
drop ball 3252 758 0.625 0.776 0.67 0.518
size measure 3252 589 0.768 0.628 0.62 0.481
record 3252 877 0.783 0.852 0.843 0.711
Speed: 0.1ms preprocess, 6.7ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/val/exp5
4.RT-DETR-r50-filter-data-enhance2
4.1.last
Ultralytics YOLOv8.0.201 🚀 Python-3.8.3 torch-1.9.0+cu111 CUDA:0 (NVIDIA A800 80GB PCIe, 80995MiB)
rtdetr-r50 summary: 480 layers, 41964383 parameters, 0 gradients, 129.6 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 425/425 [00:19<00:00, 21.31it/s]
all 3397 3832 0.701 0.692 0.685 0.515
weigh 3397 716 0.828 0.679 0.724 0.584
height measure 3397 789 0.632 0.572 0.612 0.399
drop ball 3397 830 0.59 0.717 0.613 0.434
size measure 3397 616 0.686 0.679 0.656 0.489
record 3397 881 0.768 0.814 0.822 0.666
Speed: 0.1ms preprocess, 4.3ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp7
4.2.best
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all5_filter_data_enhance/labels/val.cache... 3397 images, 0 backgrounds, 0 corrupt: 100%|██████████| 3397/3397 [00:00<?,
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 850/850 [00:28<00:00, 29.40it/s]
all 3397 3832 0.716 0.681 0.687 0.516
weigh 3397 716 0.847 0.666 0.723 0.583
height measure 3397 789 0.663 0.555 0.615 0.401
drop ball 3397 830 0.6 0.691 0.613 0.436
size measure 3397 616 0.699 0.672 0.656 0.491
record 3397 881 0.773 0.818 0.828 0.671
Speed: 0.1ms preprocess, 6.6ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/val/exp7
5.RT-DETR-r50-ASF
5.1.last
rtdetr-r50-ASF summary: 492 layers, 43080031 parameters, 0 gradients, 142.2 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 380/380 [00:19<00:00, 19.52it/s]
all 3036 3427 0.743 0.682 0.699 0.548
weigh 3036 337 0.768 0.653 0.699 0.552
height measure 3036 802 0.674 0.519 0.632 0.46
drop ball 3036 820 0.647 0.728 0.647 0.496
size measure 3036 601 0.745 0.652 0.65 0.53
record 3036 867 0.881 0.859 0.867 0.701
Speed: 0.1ms preprocess, 4.8ms inference, 0.0ms loss, 0.2ms postprocess per image
5.2.best
rtdetr-r50-ASF summary: 492 layers, 43080031 parameters, 0 gradients, 142.2 GFLOPs
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all2/labels/val.cache... 3036 images, 0 backgrounds, 0 corrupt: 100%|██████████| 3036/3036 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 759/759 [00:29<00:00, 26.04it/s]
all 3036 3427 0.742 0.683 0.699 0.548
weigh 3036 337 0.764 0.653 0.699 0.552
height measure 3036 802 0.679 0.519 0.632 0.46
drop ball 3036 820 0.644 0.73 0.644 0.494
size measure 3036 601 0.742 0.652 0.651 0.531
record 3036 867 0.883 0.859 0.869 0.704
Speed: 0.1ms preprocess, 7.6ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/val/exp3
6.RT-DETR-r50-ASF-filter
6.1.last
rtdetr-r50-ASF summary: 492 layers, 43080031 parameters, 0 gradients, 142.2 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 365/365 [00:18<00:00, 19.78it/s]
all 2915 3275 0.732 0.708 0.703 0.53
weigh 2915 337 0.674 0.694 0.678 0.538
height measure 2915 726 0.708 0.535 0.618 0.422
drop ball 2915 757 0.664 0.81 0.723 0.517
size measure 2915 589 0.795 0.644 0.644 0.479
record 2915 866 0.82 0.857 0.853 0.691
Speed: 0.1ms preprocess, 4.8ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp4
6.2.best
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all3_filter/labels/val.cache... 2915 images, 0 backgrounds, 0 corrupt: 100%|██████████| 2915/2915 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 729/729 [00:26<00:00, 27.05it/s]
all 2915 3275 0.741 0.696 0.702 0.529
weigh 2915 337 0.679 0.685 0.675 0.533
height measure 2915 726 0.729 0.522 0.62 0.423
drop ball 2915 757 0.674 0.787 0.72 0.515
size measure 2915 589 0.802 0.635 0.642 0.478
record 2915 866 0.823 0.851 0.855 0.693
Speed: 0.1ms preprocess, 7.4ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/val/exp4
7.RT-DETR-r50-ASF-filter-data-enhance
7.1.last
Ultralytics YOLOv8.0.201 🚀 Python-3.8.3 torch-1.9.0+cu111 CUDA:1 (NVIDIA A800 80GB PCIe, 80995MiB)
rtdetr-r50-ASF summary: 492 layers, 43080031 parameters, 0 gradients, 142.2 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 407/407 [00:20<00:00, 19.44it/s]
all 3252 3624 0.744 0.693 0.709 0.556
weigh 3252 674 0.825 0.709 0.742 0.609
height measure 3252 726 0.62 0.545 0.61 0.434
drop ball 3252 758 0.633 0.69 0.639 0.482
size measure 3252 589 0.794 0.664 0.67 0.527
record 3252 877 0.847 0.857 0.886 0.728
Speed: 0.1ms preprocess, 4.8ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp6
7.2.best
rtdetr-r50-ASF summary: 492 layers, 43080031 parameters, 0 gradients, 142.2 GFLOPs
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all4_data_enhance/labels/val.cache... 3252 images, 0 backgrounds, 0 corrupt: 100%|██████████| 3252/
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 813/813 [01:17<00:00, 10.50it/s]
all 3252 3624 0.744 0.702 0.711 0.557
weigh 3252 674 0.833 0.715 0.746 0.612
height measure 3252 726 0.615 0.559 0.614 0.435
drop ball 3252 758 0.634 0.706 0.638 0.48
size measure 3252 589 0.793 0.672 0.674 0.53
record 3252 877 0.847 0.86 0.887 0.729
Speed: 0.4ms preprocess, 11.5ms inference, 0.0ms loss, 0.4ms postprocess per image
Results saved to runs/val/exp6
8.RT-DETR-r50-ASF-filter-data-enhance2
8.1.last
rtdetr-r50-ASF summary: 492 layers, 43080031 parameters, 0 gradients, 142.2 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 425/425 [00:21<00:00, 19.54it/s]
all 3397 3832 0.733 0.707 0.709 0.55
weigh 3397 716 0.833 0.702 0.752 0.611
height measure 3397 789 0.619 0.561 0.602 0.438
drop ball 3397 830 0.621 0.757 0.662 0.493
size measure 3397 616 0.755 0.685 0.695 0.517
record 3397 881 0.836 0.83 0.832 0.689
Speed: 0.1ms preprocess, 4.8ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/train/exp8
8.2.best
val: Scanning /data/RT-DETR/yolo_behavior_Dataset_all5_filter_data_enhance/labels/val.cache... 3397 images, 0 backgrounds, 0 corrupt: 100%|██████████| 3397/3397 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 850/850 [00:31<00:00, 26.84it/s]
all 3397 3832 0.73 0.711 0.708 0.549
weigh 3397 716 0.834 0.7 0.752 0.609
height measure 3397 789 0.62 0.572 0.604 0.437
drop ball 3397 830 0.615 0.76 0.661 0.493
size measure 3397 616 0.748 0.689 0.693 0.515
record 3397 881 0.833 0.832 0.832 0.689
Speed: 0.1ms preprocess, 7.5ms inference, 0.0ms loss, 0.2ms postprocess per image
Results saved to runs/val/exp8