You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
model = YOLO('ultralytics/cfg/models/v8/yolov8n-pose.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-FaceRandomCrop',
)
CUDA_VISIBLE_DEVICES=1 nohup python train.py > logs/yolov8n-face-filter-small-object-FaceRandomCrop-5.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-FaceRandomCrop-5.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-FaceRandomCrop5/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
# FaceRandomCrop(max_crop_ratio=0.5, p=0.2)
==================== Results ====================
Easy Val AP: 0.9340345080128105
Medium Val AP: 0.9088230591031317
Hard Val AP: 0.7644932298164093
=================================================
# FaceRandomCrop(max_crop_ratio=0.2, p=0.5)
==================== Results ====================
Easy Val AP: 0.9294032908095071
Medium Val AP: 0.9024018857067466
Hard Val AP: 0.7410233830985312
=================================================
# FaceRandomCrop(max_crop_ratio=0.2, p=0.2)
==================== Results ====================
Easy Val AP: 0.9328363222763076
Medium Val AP: 0.9093156120152489
Hard Val AP: 0.7659403333843626
=================================================
# FaceRandomCrop(max_crop_ratio=0.1, p=0.2)
==================== Results ====================
Easy Val AP: 0.9347491736828661
Medium Val AP: 0.9101764295848018
Hard Val AP: 0.7669252759793007
=================================================
# FaceRandomCrop(max_crop_ratio=0.1, p=0.1)
==================== Results ====================
Easy Val AP: 0.9357439003883163
Medium Val AP: 0.9148017297714196
Hard Val AP: 0.7730805888976859
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL
model = YOLO('ultralytics/cfg/models/v8/yolov8n-pose.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-TAL',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-TAL-4.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-TAL-4.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-TAL4/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
# TaskAlignedAssigner(topk=7, num_classes=self.nc, alpha=0.5, beta=6.0)
==================== Results ====================
Easy Val AP: 0.9337907174770645
Medium Val AP: 0.914811616615522
Hard Val AP: 0.7820088430693088
=================================================
# TaskAlignedAssigner(topk=5, num_classes=self.nc, alpha=0.5, beta=6.0)
==================== Results ====================
Easy Val AP: 0.9349248542040435
Medium Val AP: 0.9168321887811575
Hard Val AP: 0.7865576698425549
=================================================
# TaskAlignedAssigner(topk=3, num_classes=self.nc, alpha=0.5, beta=6.0)
==================== Results ====================
Easy Val AP: 0.9308420340982722
Medium Val AP: 0.9152875519744202
Hard Val AP: 0.7872064919703559
=================================================
# TaskAlignedAssigner(topk=3, num_classes=self.nc, alpha=0.5, beta=9.0)
==================== Results ====================
Easy Val AP: 0.9256510306305094
Medium Val AP: 0.907209934965147
Hard Val AP: 0.7887162574525065
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + sppf-3
model = YOLO('ultralytics/cfg/models/v8/yolov8n-pose-sppf-3.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-sppf-3',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-sppf-3.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-sppf-3.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-sppf-3/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-sppf-3/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9322859487121787
Medium Val AP: 0.9147386945529493
Hard Val AP: 0.7872254343617164
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6',
)
CUDA_VISIBLE_DEVICES=1 nohup python train.py > logs/yolov8n-face-filter-small-object-p6 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9355275260313918
Medium Val AP: 0.9190118329407484
Hard Val AP: 0.7891111955535621
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6-C2f
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-C2f.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-C2f',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-C2f.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-C2f.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-C2f/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-C2f/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9333051410243601
Medium Val AP: 0.9155042175860136
Hard Val AP: 0.7856901170745433
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6 + ADown
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-ADown.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-ADown',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-adown 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-adown
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-ADown/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-ADown/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9336307684774567
Medium Val AP: 0.9150758849612703
Hard Val AP: 0.7704155402613113
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6 + V7Down
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-V7Down.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-V7Down',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-v7down.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-v7down.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-V7Down/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-V7Down/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9354881085278206
Medium Val AP: 0.9183207035388947
Hard Val AP: 0.7886194672127571
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6 + HGStem
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-HGStem.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-HGStem',
)
CUDA_VISIBLE_DEVICES=1 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-HGStem.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-HGStem.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-HGStem/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-HGStem/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9412578200706179
Medium Val AP: 0.9283130060394659
Hard Val AP: 0.8100328894210891
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6 + LSCD
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-LSCD.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-LSCD',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-LSCD.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-LSCD.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-LSCD/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-LSCD/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9368110776925447
Medium Val AP: 0.9195816086070555
Hard Val AP: 0.7886834195267196
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6 + HGStem + LSCD
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-HGStem-LSCD.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-HGStem-LSCD',
)
CUDA_VISIBLE_DEVICES=0 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-HGStem-LSCD.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-HGStem-LSCD.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-HGStem-LSCD/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-HGStem-LSCD/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9418271455988033
Medium Val AP: 0.9277675495505393
Hard Val AP: 0.8074121227491434
=================================================
yolov8n-pose filter 5 pixel lowprecision object in 640 images-size + TAL + P6 + HGStem + LSCD + BIFPN
model = YOLO('ultralytics/cfg/models/v8/yolov8-pose-p6-HGStem-LSCD-BIFPN.yaml')
model.train(data='data/widerface_filter_small.yaml',
cache=False,
imgsz=640,
epochs=300,
batch=16,
close_mosaic=0,
workers=8,
patience=50,
optimizer='SGD', # using SGD
project='runs/widerface',
name='yolov8n-face-filter-small-object-p6-HGStem-LSCD-BIFPN',
)
CUDA_VISIBLE_DEVICES=1 nohup python train.py > logs/yolov8n-face-filter-small-object-p6-HGStem-LSCD-BIFPN.log 2>&1 & tail -f logs/yolov8n-face-filter-small-object-p6-HGStem-LSCD-BIFPN.log
CUDA_VISIBLE_DEVICES=0 python test_widerface.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-HGStem-LSCD-BIFPN/weights/best.pt --dataset_folder /root/data_ssd/WIDER-FACE/WIDER_val/images --num_workers 8
python get_inference_time.py --weights runs/widerface/yolov8n-face-filter-small-object-p6-HGStem-LSCD-BIFPN/weights/last.pt --warmup 100 --testtime 300 --batch 32 --device 0
==================== Results ====================
Easy Val AP: 0.9419754472002038
Medium Val AP: 0.9289276822471543
Hard Val AP: 0.8114986117385863
=================================================