glenn-jocher
commited on
Commit
β’
4821d07
1
Parent(s):
d3e7778
Increment train, test, detect runs/ (#1322)
Browse files* Increment train, test, detect runs/
* Update ci-testing.yml
* inference/images to data/images
* move images
* runs/exp to runs/train/exp
* update 'results saved to %s' str
- .github/workflows/ci-testing.yml +2 -2
- .gitignore +2 -2
- Dockerfile +1 -1
- README.md +7 -7
- {inference β data}/images/bus.jpg +0 -0
- {inference β data}/images/zidane.jpg +0 -0
- detect.py +20 -18
- hubconf.py +1 -1
- sotabench.py +0 -307
- test.py +12 -18
- train.py +5 -6
- tutorial.ipynb +18 -18
- utils/general.py +8 -2
.github/workflows/ci-testing.yml
CHANGED
@@ -66,10 +66,10 @@ jobs:
|
|
66 |
python train.py --img 256 --batch 8 --weights weights/${{ matrix.model }}.pt --cfg models/${{ matrix.model }}.yaml --epochs 1 --device $di
|
67 |
# detect
|
68 |
python detect.py --weights weights/${{ matrix.model }}.pt --device $di
|
69 |
-
python detect.py --weights runs/exp0/weights/last.pt --device $di
|
70 |
# test
|
71 |
python test.py --img 256 --batch 8 --weights weights/${{ matrix.model }}.pt --device $di
|
72 |
-
python test.py --img 256 --batch 8 --weights runs/exp0/weights/last.pt --device $di
|
73 |
|
74 |
python models/yolo.py --cfg models/${{ matrix.model }}.yaml # inspect
|
75 |
python models/export.py --img 256 --batch 1 --weights weights/${{ matrix.model }}.pt # export
|
|
|
66 |
python train.py --img 256 --batch 8 --weights weights/${{ matrix.model }}.pt --cfg models/${{ matrix.model }}.yaml --epochs 1 --device $di
|
67 |
# detect
|
68 |
python detect.py --weights weights/${{ matrix.model }}.pt --device $di
|
69 |
+
python detect.py --weights runs/train/exp0/weights/last.pt --device $di
|
70 |
# test
|
71 |
python test.py --img 256 --batch 8 --weights weights/${{ matrix.model }}.pt --device $di
|
72 |
+
python test.py --img 256 --batch 8 --weights runs/train/exp0/weights/last.pt --device $di
|
73 |
|
74 |
python models/yolo.py --cfg models/${{ matrix.model }}.yaml # inspect
|
75 |
python models/export.py --img 256 --batch 1 --weights weights/${{ matrix.model }}.pt # export
|
.gitignore
CHANGED
@@ -26,8 +26,8 @@
|
|
26 |
storage.googleapis.com
|
27 |
runs/*
|
28 |
data/*
|
29 |
-
!data/
|
30 |
-
!data/
|
31 |
!data/coco.names
|
32 |
!data/coco_paper.names
|
33 |
!data/coco.data
|
|
|
26 |
storage.googleapis.com
|
27 |
runs/*
|
28 |
data/*
|
29 |
+
!data/images/zidane.jpg
|
30 |
+
!data/images/bus.jpg
|
31 |
!data/coco.names
|
32 |
!data/coco_paper.names
|
33 |
!data/coco.data
|
Dockerfile
CHANGED
@@ -46,7 +46,7 @@ COPY . /usr/src/app
|
|
46 |
# sudo docker commit 092b16b25c5b usr/resume && sudo docker run -it --gpus all --ipc=host -v "$(pwd)"/coco:/usr/src/coco --entrypoint=sh usr/resume
|
47 |
|
48 |
# Send weights to GCP
|
49 |
-
# python -c "from utils.general import *; strip_optimizer('runs/exp0_*/weights/best.pt', 'tmp.pt')" && gsutil cp tmp.pt gs://*.pt
|
50 |
|
51 |
# Clean up
|
52 |
# docker system prune -a --volumes
|
|
|
46 |
# sudo docker commit 092b16b25c5b usr/resume && sudo docker run -it --gpus all --ipc=host -v "$(pwd)"/coco:/usr/src/coco --entrypoint=sh usr/resume
|
47 |
|
48 |
# Send weights to GCP
|
49 |
+
# python -c "from utils.general import *; strip_optimizer('runs/train/exp0_*/weights/best.pt', 'tmp.pt')" && gsutil cp tmp.pt gs://*.pt
|
50 |
|
51 |
# Clean up
|
52 |
# docker system prune -a --volumes
|
README.md
CHANGED
@@ -70,7 +70,7 @@ YOLOv5 may be run in any of the following up-to-date verified environments (with
|
|
70 |
|
71 |
## Inference
|
72 |
|
73 |
-
detect.py runs inference on a variety of sources, downloading models automatically from the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases) and saving results to `
|
74 |
```bash
|
75 |
$ python detect.py --source 0 # webcam
|
76 |
file.jpg # image
|
@@ -82,20 +82,20 @@ $ python detect.py --source 0 # webcam
|
|
82 |
http://112.50.243.8/PLTV/88888888/224/3221225900/1.m3u8 # http stream
|
83 |
```
|
84 |
|
85 |
-
To run inference on example images in `
|
86 |
```bash
|
87 |
-
$ python detect.py --source
|
88 |
|
89 |
-
Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.25, device='', img_size=640, iou_thres=0.45, output='
|
90 |
Using CUDA device0 _CudaDeviceProperties(name='Tesla V100-SXM2-16GB', total_memory=16160MB)
|
91 |
|
92 |
Downloading https://github.com/ultralytics/yolov5/releases/download/v3.0/yolov5s.pt to yolov5s.pt... 100%|ββββββββββββββ| 14.5M/14.5M [00:00<00:00, 21.3MB/s]
|
93 |
|
94 |
Fusing layers...
|
95 |
Model Summary: 140 layers, 7.45958e+06 parameters, 0 gradients
|
96 |
-
image 1/2
|
97 |
-
image 2/2
|
98 |
-
Results saved to
|
99 |
Done. (0.124s)
|
100 |
```
|
101 |
<img src="https://user-images.githubusercontent.com/26833433/97107365-685a8d80-16c7-11eb-8c2e-83aac701d8b9.jpeg" width="500">
|
|
|
70 |
|
71 |
## Inference
|
72 |
|
73 |
+
detect.py runs inference on a variety of sources, downloading models automatically from the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases) and saving results to `runs/detect`.
|
74 |
```bash
|
75 |
$ python detect.py --source 0 # webcam
|
76 |
file.jpg # image
|
|
|
82 |
http://112.50.243.8/PLTV/88888888/224/3221225900/1.m3u8 # http stream
|
83 |
```
|
84 |
|
85 |
+
To run inference on example images in `data/images`:
|
86 |
```bash
|
87 |
+
$ python detect.py --source data/images --weights yolov5s.pt --conf 0.25
|
88 |
|
89 |
+
Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.25, device='', img_size=640, iou_thres=0.45, output='runs/detect', save_conf=False, save_txt=False, source='data/images', update=False, view_img=False, weights='yolov5s.pt')
|
90 |
Using CUDA device0 _CudaDeviceProperties(name='Tesla V100-SXM2-16GB', total_memory=16160MB)
|
91 |
|
92 |
Downloading https://github.com/ultralytics/yolov5/releases/download/v3.0/yolov5s.pt to yolov5s.pt... 100%|ββββββββββββββ| 14.5M/14.5M [00:00<00:00, 21.3MB/s]
|
93 |
|
94 |
Fusing layers...
|
95 |
Model Summary: 140 layers, 7.45958e+06 parameters, 0 gradients
|
96 |
+
image 1/2 data/images/bus.jpg: 640x480 4 persons, 1 buss, 1 skateboards, Done. (0.013s)
|
97 |
+
image 2/2 data/images/zidane.jpg: 384x640 2 persons, 2 ties, Done. (0.013s)
|
98 |
+
Results saved to runs/detect/exp0
|
99 |
Done. (0.124s)
|
100 |
```
|
101 |
<img src="https://user-images.githubusercontent.com/26833433/97107365-685a8d80-16c7-11eb-8c2e-83aac701d8b9.jpeg" width="500">
|
{inference β data}/images/bus.jpg
RENAMED
File without changes
|
{inference β data}/images/zidane.jpg
RENAMED
File without changes
|
detect.py
CHANGED
@@ -1,6 +1,5 @@
|
|
1 |
import argparse
|
2 |
import os
|
3 |
-
import shutil
|
4 |
import time
|
5 |
from pathlib import Path
|
6 |
|
@@ -11,23 +10,25 @@ from numpy import random
|
|
11 |
|
12 |
from models.experimental import attempt_load
|
13 |
from utils.datasets import LoadStreams, LoadImages
|
14 |
-
from utils.general import
|
15 |
-
|
16 |
-
xyxy2xywh, plot_one_box, strip_optimizer, set_logging)
|
17 |
from utils.torch_utils import select_device, load_classifier, time_synchronized
|
18 |
|
19 |
|
20 |
def detect(save_img=False):
|
21 |
-
|
22 |
-
opt.save_dir, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
|
23 |
webcam = source.isnumeric() or source.startswith(('rtsp://', 'rtmp://', 'http://')) or source.endswith('.txt')
|
24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
# Initialize
|
26 |
set_logging()
|
27 |
device = select_device(opt.device)
|
28 |
-
if os.path.exists(out): # output dir
|
29 |
-
shutil.rmtree(out) # delete dir
|
30 |
-
os.makedirs(out) # make new dir
|
31 |
half = device.type != 'cpu' # half precision only supported on CUDA
|
32 |
|
33 |
# Load model
|
@@ -83,12 +84,12 @@ def detect(save_img=False):
|
|
83 |
# Process detections
|
84 |
for i, det in enumerate(pred): # detections per image
|
85 |
if webcam: # batch_size >= 1
|
86 |
-
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
|
87 |
else:
|
88 |
-
p, s, im0 = path, '', im0s
|
89 |
|
90 |
-
save_path = str(
|
91 |
-
txt_path = str(
|
92 |
s += '%gx%g ' % img.shape[2:] # print string
|
93 |
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
|
94 |
if det is not None and len(det):
|
@@ -104,7 +105,7 @@ def detect(save_img=False):
|
|
104 |
for *xyxy, conf, cls in reversed(det):
|
105 |
if save_txt: # Write to file
|
106 |
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
107 |
-
line = (cls,
|
108 |
with open(txt_path + '.txt', 'a') as f:
|
109 |
f.write(('%g ' * len(line) + '\n') % line)
|
110 |
|
@@ -139,7 +140,7 @@ def detect(save_img=False):
|
|
139 |
vid_writer.write(im0)
|
140 |
|
141 |
if save_txt or save_img:
|
142 |
-
print('Results saved to %s' %
|
143 |
|
144 |
print('Done. (%.3fs)' % (time.time() - t0))
|
145 |
|
@@ -147,15 +148,16 @@ def detect(save_img=False):
|
|
147 |
if __name__ == '__main__':
|
148 |
parser = argparse.ArgumentParser()
|
149 |
parser.add_argument('--weights', nargs='+', type=str, default='yolov5s.pt', help='model.pt path(s)')
|
150 |
-
parser.add_argument('--source', type=str, default='
|
151 |
parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)')
|
152 |
parser.add_argument('--conf-thres', type=float, default=0.25, help='object confidence threshold')
|
153 |
parser.add_argument('--iou-thres', type=float, default=0.45, help='IOU threshold for NMS')
|
154 |
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
155 |
parser.add_argument('--view-img', action='store_true', help='display results')
|
156 |
-
parser.add_argument('--save-txt', action='
|
157 |
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
|
158 |
-
parser.add_argument('--save-dir', type=str, default='
|
|
|
159 |
parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3')
|
160 |
parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
|
161 |
parser.add_argument('--augment', action='store_true', help='augmented inference')
|
|
|
1 |
import argparse
|
2 |
import os
|
|
|
3 |
import time
|
4 |
from pathlib import Path
|
5 |
|
|
|
10 |
|
11 |
from models.experimental import attempt_load
|
12 |
from utils.datasets import LoadStreams, LoadImages
|
13 |
+
from utils.general import check_img_size, non_max_suppression, apply_classifier, scale_coords, xyxy2xywh, \
|
14 |
+
plot_one_box, strip_optimizer, set_logging, increment_dir
|
|
|
15 |
from utils.torch_utils import select_device, load_classifier, time_synchronized
|
16 |
|
17 |
|
18 |
def detect(save_img=False):
|
19 |
+
save_dir, source, weights, view_img, save_txt, imgsz = \
|
20 |
+
Path(opt.save_dir), opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
|
21 |
webcam = source.isnumeric() or source.startswith(('rtsp://', 'rtmp://', 'http://')) or source.endswith('.txt')
|
22 |
|
23 |
+
# Directories
|
24 |
+
if save_dir == Path('runs/detect'): # if default
|
25 |
+
os.makedirs('runs/detect', exist_ok=True) # make base
|
26 |
+
save_dir = Path(increment_dir(save_dir / 'exp', opt.name)) # increment run
|
27 |
+
os.makedirs(save_dir / 'labels' if save_txt else save_dir, exist_ok=True) # make new dir
|
28 |
+
|
29 |
# Initialize
|
30 |
set_logging()
|
31 |
device = select_device(opt.device)
|
|
|
|
|
|
|
32 |
half = device.type != 'cpu' # half precision only supported on CUDA
|
33 |
|
34 |
# Load model
|
|
|
84 |
# Process detections
|
85 |
for i, det in enumerate(pred): # detections per image
|
86 |
if webcam: # batch_size >= 1
|
87 |
+
p, s, im0 = Path(path[i]), '%g: ' % i, im0s[i].copy()
|
88 |
else:
|
89 |
+
p, s, im0 = Path(path), '', im0s
|
90 |
|
91 |
+
save_path = str(save_dir / p.name)
|
92 |
+
txt_path = str(save_dir / 'labels' / p.stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
|
93 |
s += '%gx%g ' % img.shape[2:] # print string
|
94 |
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
|
95 |
if det is not None and len(det):
|
|
|
105 |
for *xyxy, conf, cls in reversed(det):
|
106 |
if save_txt: # Write to file
|
107 |
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
108 |
+
line = (cls, *xywh, conf) if opt.save_conf else (cls, *xywh) # label format
|
109 |
with open(txt_path + '.txt', 'a') as f:
|
110 |
f.write(('%g ' * len(line) + '\n') % line)
|
111 |
|
|
|
140 |
vid_writer.write(im0)
|
141 |
|
142 |
if save_txt or save_img:
|
143 |
+
print('Results saved to %s' % save_dir)
|
144 |
|
145 |
print('Done. (%.3fs)' % (time.time() - t0))
|
146 |
|
|
|
148 |
if __name__ == '__main__':
|
149 |
parser = argparse.ArgumentParser()
|
150 |
parser.add_argument('--weights', nargs='+', type=str, default='yolov5s.pt', help='model.pt path(s)')
|
151 |
+
parser.add_argument('--source', type=str, default='data/images', help='source') # file/folder, 0 for webcam
|
152 |
parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)')
|
153 |
parser.add_argument('--conf-thres', type=float, default=0.25, help='object confidence threshold')
|
154 |
parser.add_argument('--iou-thres', type=float, default=0.45, help='IOU threshold for NMS')
|
155 |
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
156 |
parser.add_argument('--view-img', action='store_true', help='display results')
|
157 |
+
parser.add_argument('--save-txt', action='store_false', help='save results to *.txt')
|
158 |
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
|
159 |
+
parser.add_argument('--save-dir', type=str, default='runs/detect', help='directory to save results')
|
160 |
+
parser.add_argument('--name', default='', help='name to append to --save-dir: i.e. runs/{N} -> runs/{N}_{name}')
|
161 |
parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3')
|
162 |
parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
|
163 |
parser.add_argument('--augment', action='store_true', help='augmented inference')
|
hubconf.py
CHANGED
@@ -113,6 +113,6 @@ if __name__ == '__main__':
|
|
113 |
# Verify inference
|
114 |
from PIL import Image
|
115 |
|
116 |
-
img = Image.open('
|
117 |
y = model(img)
|
118 |
print(y[0].shape)
|
|
|
113 |
# Verify inference
|
114 |
from PIL import Image
|
115 |
|
116 |
+
img = Image.open('data/images/zidane.jpg')
|
117 |
y = model(img)
|
118 |
print(y[0].shape)
|
sotabench.py
DELETED
@@ -1,307 +0,0 @@
|
|
1 |
-
import argparse
|
2 |
-
import glob
|
3 |
-
import os
|
4 |
-
import shutil
|
5 |
-
from pathlib import Path
|
6 |
-
|
7 |
-
import numpy as np
|
8 |
-
import torch
|
9 |
-
import yaml
|
10 |
-
from sotabencheval.object_detection import COCOEvaluator
|
11 |
-
from sotabencheval.utils import is_server
|
12 |
-
from tqdm import tqdm
|
13 |
-
|
14 |
-
from models.experimental import attempt_load
|
15 |
-
from utils.datasets import create_dataloader
|
16 |
-
from utils.general import (
|
17 |
-
coco80_to_coco91_class, check_dataset, check_file, check_img_size, compute_loss, non_max_suppression, scale_coords,
|
18 |
-
xyxy2xywh, clip_coords, set_logging)
|
19 |
-
from utils.torch_utils import select_device, time_synchronized
|
20 |
-
|
21 |
-
DATA_ROOT = './.data/vision/coco' if is_server() else '../coco' # sotabench data dir
|
22 |
-
|
23 |
-
|
24 |
-
def test(data,
|
25 |
-
weights=None,
|
26 |
-
batch_size=16,
|
27 |
-
imgsz=640,
|
28 |
-
conf_thres=0.001,
|
29 |
-
iou_thres=0.6, # for NMS
|
30 |
-
save_json=False,
|
31 |
-
single_cls=False,
|
32 |
-
augment=False,
|
33 |
-
verbose=False,
|
34 |
-
model=None,
|
35 |
-
dataloader=None,
|
36 |
-
save_dir='',
|
37 |
-
merge=False,
|
38 |
-
save_txt=False):
|
39 |
-
# Initialize/load model and set device
|
40 |
-
training = model is not None
|
41 |
-
if training: # called by train.py
|
42 |
-
device = next(model.parameters()).device # get model device
|
43 |
-
|
44 |
-
else: # called directly
|
45 |
-
set_logging()
|
46 |
-
device = select_device(opt.device, batch_size=batch_size)
|
47 |
-
merge, save_txt = opt.merge, opt.save_txt # use Merge NMS, save *.txt labels
|
48 |
-
if save_txt:
|
49 |
-
out = Path('inference/output')
|
50 |
-
if os.path.exists(out):
|
51 |
-
shutil.rmtree(out) # delete output folder
|
52 |
-
os.makedirs(out) # make new output folder
|
53 |
-
|
54 |
-
# Remove previous
|
55 |
-
for f in glob.glob(str(Path(save_dir) / 'test_batch*.jpg')):
|
56 |
-
os.remove(f)
|
57 |
-
|
58 |
-
# Load model
|
59 |
-
model = attempt_load(weights, map_location=device) # load FP32 model
|
60 |
-
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
|
61 |
-
|
62 |
-
# Multi-GPU disabled, incompatible with .half() https://github.com/ultralytics/yolov5/issues/99
|
63 |
-
# if device.type != 'cpu' and torch.cuda.device_count() > 1:
|
64 |
-
# model = nn.DataParallel(model)
|
65 |
-
|
66 |
-
# Half
|
67 |
-
half = device.type != 'cpu' # half precision only supported on CUDA
|
68 |
-
if half:
|
69 |
-
model.half()
|
70 |
-
|
71 |
-
# Configure
|
72 |
-
model.eval()
|
73 |
-
with open(data) as f:
|
74 |
-
data = yaml.load(f, Loader=yaml.FullLoader) # model dict
|
75 |
-
check_dataset(data) # check
|
76 |
-
nc = 1 if single_cls else int(data['nc']) # number of classes
|
77 |
-
iouv = torch.linspace(0.5, 0.95, 10).to(device) # iou vector for mAP@0.5:0.95
|
78 |
-
niou = iouv.numel()
|
79 |
-
|
80 |
-
# Dataloader
|
81 |
-
if not training:
|
82 |
-
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
|
83 |
-
_ = model(img.half() if half else img) if device.type != 'cpu' else None # run once
|
84 |
-
path = data['test'] if opt.task == 'test' else data['val'] # path to val/test images
|
85 |
-
dataloader = create_dataloader(path, imgsz, batch_size, model.stride.max(), opt,
|
86 |
-
hyp=None, augment=False, cache=True, pad=0.5, rect=True)[0]
|
87 |
-
|
88 |
-
seen = 0
|
89 |
-
names = model.names if hasattr(model, 'names') else model.module.names
|
90 |
-
coco91class = coco80_to_coco91_class()
|
91 |
-
s = ('%20s' + '%12s' * 6) % ('Class', 'Images', 'Targets', 'P', 'R', 'mAP@.5', 'mAP@.5:.95')
|
92 |
-
p, r, f1, mp, mr, map50, map, t0, t1 = 0., 0., 0., 0., 0., 0., 0., 0., 0.
|
93 |
-
loss = torch.zeros(3, device=device)
|
94 |
-
jdict, stats, ap, ap_class = [], [], [], []
|
95 |
-
evaluator = COCOEvaluator(root=DATA_ROOT, model_name=opt.weights.replace('.pt', ''))
|
96 |
-
for batch_i, (img, targets, paths, shapes) in enumerate(tqdm(dataloader, desc=s)):
|
97 |
-
img = img.to(device, non_blocking=True)
|
98 |
-
img = img.half() if half else img.float() # uint8 to fp16/32
|
99 |
-
img /= 255.0 # 0 - 255 to 0.0 - 1.0
|
100 |
-
targets = targets.to(device)
|
101 |
-
nb, _, height, width = img.shape # batch size, channels, height, width
|
102 |
-
whwh = torch.Tensor([width, height, width, height]).to(device)
|
103 |
-
|
104 |
-
# Disable gradients
|
105 |
-
with torch.no_grad():
|
106 |
-
# Run model
|
107 |
-
t = time_synchronized()
|
108 |
-
inf_out, train_out = model(img, augment=augment) # inference and training outputs
|
109 |
-
t0 += time_synchronized() - t
|
110 |
-
|
111 |
-
# Compute loss
|
112 |
-
if training: # if model has loss hyperparameters
|
113 |
-
loss += compute_loss([x.float() for x in train_out], targets, model)[1][:3] # box, obj, cls
|
114 |
-
|
115 |
-
# Run NMS
|
116 |
-
t = time_synchronized()
|
117 |
-
output = non_max_suppression(inf_out, conf_thres=conf_thres, iou_thres=iou_thres, merge=merge)
|
118 |
-
t1 += time_synchronized() - t
|
119 |
-
|
120 |
-
# Statistics per image
|
121 |
-
for si, pred in enumerate(output):
|
122 |
-
labels = targets[targets[:, 0] == si, 1:]
|
123 |
-
nl = len(labels)
|
124 |
-
tcls = labels[:, 0].tolist() if nl else [] # target class
|
125 |
-
seen += 1
|
126 |
-
|
127 |
-
if pred is None:
|
128 |
-
if nl:
|
129 |
-
stats.append((torch.zeros(0, niou, dtype=torch.bool), torch.Tensor(), torch.Tensor(), tcls))
|
130 |
-
continue
|
131 |
-
|
132 |
-
# Append to text file
|
133 |
-
if save_txt:
|
134 |
-
gn = torch.tensor(shapes[si][0])[[1, 0, 1, 0]] # normalization gain whwh
|
135 |
-
x = pred.clone()
|
136 |
-
x[:, :4] = scale_coords(img[si].shape[1:], x[:, :4], shapes[si][0], shapes[si][1]) # to original
|
137 |
-
for *xyxy, conf, cls in x:
|
138 |
-
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
139 |
-
with open(str(out / Path(paths[si]).stem) + '.txt', 'a') as f:
|
140 |
-
f.write(('%g ' * 5 + '\n') % (cls, *xywh)) # label format
|
141 |
-
|
142 |
-
# Clip boxes to image bounds
|
143 |
-
clip_coords(pred, (height, width))
|
144 |
-
|
145 |
-
# Append to pycocotools JSON dictionary
|
146 |
-
if save_json:
|
147 |
-
# [{"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}, ...
|
148 |
-
image_id = Path(paths[si]).stem
|
149 |
-
box = pred[:, :4].clone() # xyxy
|
150 |
-
scale_coords(img[si].shape[1:], box, shapes[si][0], shapes[si][1]) # to original shape
|
151 |
-
box = xyxy2xywh(box) # xywh
|
152 |
-
box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner
|
153 |
-
for p, b in zip(pred.tolist(), box.tolist()):
|
154 |
-
result = {'image_id': int(image_id) if image_id.isnumeric() else image_id,
|
155 |
-
'category_id': coco91class[int(p[5])],
|
156 |
-
'bbox': [round(x, 3) for x in b],
|
157 |
-
'score': round(p[4], 5)}
|
158 |
-
jdict.append(result)
|
159 |
-
|
160 |
-
#evaluator.add([result])
|
161 |
-
#if evaluator.cache_exists:
|
162 |
-
# break
|
163 |
-
|
164 |
-
# # Assign all predictions as incorrect
|
165 |
-
# correct = torch.zeros(pred.shape[0], niou, dtype=torch.bool, device=device)
|
166 |
-
# if nl:
|
167 |
-
# detected = [] # target indices
|
168 |
-
# tcls_tensor = labels[:, 0]
|
169 |
-
#
|
170 |
-
# # target boxes
|
171 |
-
# tbox = xywh2xyxy(labels[:, 1:5]) * whwh
|
172 |
-
#
|
173 |
-
# # Per target class
|
174 |
-
# for cls in torch.unique(tcls_tensor):
|
175 |
-
# ti = (cls == tcls_tensor).nonzero(as_tuple=False).view(-1) # prediction indices
|
176 |
-
# pi = (cls == pred[:, 5]).nonzero(as_tuple=False).view(-1) # target indices
|
177 |
-
#
|
178 |
-
# # Search for detections
|
179 |
-
# if pi.shape[0]:
|
180 |
-
# # Prediction to target ious
|
181 |
-
# ious, i = box_iou(pred[pi, :4], tbox[ti]).max(1) # best ious, indices
|
182 |
-
#
|
183 |
-
# # Append detections
|
184 |
-
# detected_set = set()
|
185 |
-
# for j in (ious > iouv[0]).nonzero(as_tuple=False):
|
186 |
-
# d = ti[i[j]] # detected target
|
187 |
-
# if d.item() not in detected_set:
|
188 |
-
# detected_set.add(d.item())
|
189 |
-
# detected.append(d)
|
190 |
-
# correct[pi[j]] = ious[j] > iouv # iou_thres is 1xn
|
191 |
-
# if len(detected) == nl: # all targets already located in image
|
192 |
-
# break
|
193 |
-
#
|
194 |
-
# # Append statistics (correct, conf, pcls, tcls)
|
195 |
-
# stats.append((correct.cpu(), pred[:, 4].cpu(), pred[:, 5].cpu(), tcls))
|
196 |
-
|
197 |
-
# # Plot images
|
198 |
-
# if batch_i < 1:
|
199 |
-
# f = Path(save_dir) / ('test_batch%g_gt.jpg' % batch_i) # filename
|
200 |
-
# plot_images(img, targets, paths, str(f), names) # ground truth
|
201 |
-
# f = Path(save_dir) / ('test_batch%g_pred.jpg' % batch_i)
|
202 |
-
# plot_images(img, output_to_target(output, width, height), paths, str(f), names) # predictions
|
203 |
-
|
204 |
-
evaluator.add(jdict)
|
205 |
-
evaluator.save()
|
206 |
-
|
207 |
-
# # Compute statistics
|
208 |
-
# stats = [np.concatenate(x, 0) for x in zip(*stats)] # to numpy
|
209 |
-
# if len(stats) and stats[0].any():
|
210 |
-
# p, r, ap, f1, ap_class = ap_per_class(*stats)
|
211 |
-
# p, r, ap50, ap = p[:, 0], r[:, 0], ap[:, 0], ap.mean(1) # [P, R, AP@0.5, AP@0.5:0.95]
|
212 |
-
# mp, mr, map50, map = p.mean(), r.mean(), ap50.mean(), ap.mean()
|
213 |
-
# nt = np.bincount(stats[3].astype(np.int64), minlength=nc) # number of targets per class
|
214 |
-
# else:
|
215 |
-
# nt = torch.zeros(1)
|
216 |
-
#
|
217 |
-
# # Print results
|
218 |
-
# pf = '%20s' + '%12.3g' * 6 # print format
|
219 |
-
# print(pf % ('all', seen, nt.sum(), mp, mr, map50, map))
|
220 |
-
#
|
221 |
-
# # Print results per class
|
222 |
-
# if verbose and nc > 1 and len(stats):
|
223 |
-
# for i, c in enumerate(ap_class):
|
224 |
-
# print(pf % (names[c], seen, nt[c], p[i], r[i], ap50[i], ap[i]))
|
225 |
-
#
|
226 |
-
# # Print speeds
|
227 |
-
# t = tuple(x / seen * 1E3 for x in (t0, t1, t0 + t1)) + (imgsz, imgsz, batch_size) # tuple
|
228 |
-
# if not training:
|
229 |
-
# print('Speed: %.1f/%.1f/%.1f ms inference/NMS/total per %gx%g image at batch-size %g' % t)
|
230 |
-
#
|
231 |
-
# # Save JSON
|
232 |
-
# if save_json and len(jdict):
|
233 |
-
# f = 'detections_val2017_%s_results.json' % \
|
234 |
-
# (weights.split(os.sep)[-1].replace('.pt', '') if isinstance(weights, str) else '') # filename
|
235 |
-
# print('\nCOCO mAP with pycocotools... saving %s...' % f)
|
236 |
-
# with open(f, 'w') as file:
|
237 |
-
# json.dump(jdict, file)
|
238 |
-
#
|
239 |
-
# try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
|
240 |
-
# from pycocotools.coco import COCO
|
241 |
-
# from pycocotools.cocoeval import COCOeval
|
242 |
-
#
|
243 |
-
# imgIds = [int(Path(x).stem) for x in dataloader.dataset.img_files]
|
244 |
-
# cocoGt = COCO(glob.glob('../coco/annotations/instances_val*.json')[0]) # initialize COCO ground truth api
|
245 |
-
# cocoDt = cocoGt.loadRes(f) # initialize COCO pred api
|
246 |
-
# cocoEval = COCOeval(cocoGt, cocoDt, 'bbox')
|
247 |
-
# cocoEval.params.imgIds = imgIds # image IDs to evaluate
|
248 |
-
# cocoEval.evaluate()
|
249 |
-
# cocoEval.accumulate()
|
250 |
-
# cocoEval.summarize()
|
251 |
-
# map, map50 = cocoEval.stats[:2] # update results (mAP@0.5:0.95, mAP@0.5)
|
252 |
-
# except Exception as e:
|
253 |
-
# print('ERROR: pycocotools unable to run: %s' % e)
|
254 |
-
#
|
255 |
-
# # Return results
|
256 |
-
# model.float() # for training
|
257 |
-
# maps = np.zeros(nc) + map
|
258 |
-
# for i, c in enumerate(ap_class):
|
259 |
-
# maps[c] = ap[i]
|
260 |
-
# return (mp, mr, map50, map, *(loss.cpu() / len(dataloader)).tolist()), maps, t
|
261 |
-
|
262 |
-
|
263 |
-
if __name__ == '__main__':
|
264 |
-
parser = argparse.ArgumentParser(prog='test.py')
|
265 |
-
parser.add_argument('--weights', nargs='+', type=str, default='yolov5s.pt', help='model.pt path(s)')
|
266 |
-
parser.add_argument('--data', type=str, default='data/coco.yaml', help='*.data path')
|
267 |
-
parser.add_argument('--batch-size', type=int, default=32, help='size of each image batch')
|
268 |
-
parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)')
|
269 |
-
parser.add_argument('--conf-thres', type=float, default=0.001, help='object confidence threshold')
|
270 |
-
parser.add_argument('--iou-thres', type=float, default=0.65, help='IOU threshold for NMS')
|
271 |
-
parser.add_argument('--save-json', action='store_true', help='save a cocoapi-compatible JSON results file')
|
272 |
-
parser.add_argument('--task', default='val', help="'val', 'test', 'study'")
|
273 |
-
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
274 |
-
parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset')
|
275 |
-
parser.add_argument('--augment', action='store_true', help='augmented inference')
|
276 |
-
parser.add_argument('--merge', action='store_true', help='use Merge NMS')
|
277 |
-
parser.add_argument('--verbose', action='store_true', help='report mAP by class')
|
278 |
-
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
|
279 |
-
opt = parser.parse_args()
|
280 |
-
opt.save_json |= opt.data.endswith('coco.yaml')
|
281 |
-
opt.data = check_file(opt.data) # check file
|
282 |
-
print(opt)
|
283 |
-
|
284 |
-
if opt.task in ['val', 'test']: # run normally
|
285 |
-
test(opt.data,
|
286 |
-
opt.weights,
|
287 |
-
opt.batch_size,
|
288 |
-
opt.img_size,
|
289 |
-
opt.conf_thres,
|
290 |
-
opt.iou_thres,
|
291 |
-
opt.save_json,
|
292 |
-
opt.single_cls,
|
293 |
-
opt.augment,
|
294 |
-
opt.verbose)
|
295 |
-
|
296 |
-
elif opt.task == 'study': # run over a range of settings and save/plot
|
297 |
-
for weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']:
|
298 |
-
f = 'study_%s_%s.txt' % (Path(opt.data).stem, Path(weights).stem) # filename to save to
|
299 |
-
x = list(range(320, 800, 64)) # x axis
|
300 |
-
y = [] # y axis
|
301 |
-
for i in x: # img-size
|
302 |
-
print('\nRunning %s point %s...' % (f, i))
|
303 |
-
r, _, t = test(opt.data, weights, opt.batch_size, i, opt.conf_thres, opt.iou_thres, opt.save_json)
|
304 |
-
y.append(r + t) # results and times
|
305 |
-
np.savetxt(f, y, fmt='%10.4g') # save
|
306 |
-
os.system('zip -r study.zip study_*.txt')
|
307 |
-
# utils.general.plot_study_txt(f, x) # plot
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
test.py
CHANGED
@@ -2,7 +2,6 @@ import argparse
|
|
2 |
import glob
|
3 |
import json
|
4 |
import os
|
5 |
-
import shutil
|
6 |
from pathlib import Path
|
7 |
|
8 |
import numpy as np
|
@@ -12,9 +11,9 @@ from tqdm import tqdm
|
|
12 |
|
13 |
from models.experimental import attempt_load
|
14 |
from utils.datasets import create_dataloader
|
15 |
-
from utils.general import
|
16 |
-
|
17 |
-
|
18 |
from utils.torch_utils import select_device, time_synchronized
|
19 |
|
20 |
|
@@ -46,16 +45,11 @@ def test(data,
|
|
46 |
device = select_device(opt.device, batch_size=batch_size)
|
47 |
save_txt = opt.save_txt # save *.txt labels
|
48 |
|
49 |
-
#
|
50 |
-
if
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
if save_txt:
|
55 |
-
out = save_dir / 'autolabels'
|
56 |
-
if os.path.exists(out):
|
57 |
-
shutil.rmtree(out) # delete dir
|
58 |
-
os.makedirs(out) # make new dir
|
59 |
|
60 |
# Load model
|
61 |
model = attempt_load(weights, map_location=device) # load FP32 model
|
@@ -144,8 +138,8 @@ def test(data,
|
|
144 |
x[:, :4] = scale_coords(img[si].shape[1:], x[:, :4], shapes[si][0], shapes[si][1]) # to original
|
145 |
for *xyxy, conf, cls in x:
|
146 |
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
147 |
-
line = (cls,
|
148 |
-
with open(str(
|
149 |
f.write(('%g ' * len(line) + '\n') % line)
|
150 |
|
151 |
# W&B logging
|
@@ -268,6 +262,7 @@ def test(data,
|
|
268 |
print('ERROR: pycocotools unable to run: %s' % e)
|
269 |
|
270 |
# Return results
|
|
|
271 |
model.float() # for training
|
272 |
maps = np.zeros(nc) + map
|
273 |
for i, c in enumerate(ap_class):
|
@@ -292,6 +287,7 @@ if __name__ == '__main__':
|
|
292 |
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
|
293 |
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
|
294 |
parser.add_argument('--save-dir', type=str, default='runs/test', help='directory to save results')
|
|
|
295 |
opt = parser.parse_args()
|
296 |
opt.save_json |= opt.data.endswith('coco.yaml')
|
297 |
opt.data = check_file(opt.data) # check file
|
@@ -313,8 +309,6 @@ if __name__ == '__main__':
|
|
313 |
save_conf=opt.save_conf,
|
314 |
)
|
315 |
|
316 |
-
print('Results saved to %s' % opt.save_dir)
|
317 |
-
|
318 |
elif opt.task == 'study': # run over a range of settings and save/plot
|
319 |
for weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']:
|
320 |
f = 'study_%s_%s.txt' % (Path(opt.data).stem, Path(weights).stem) # filename to save to
|
|
|
2 |
import glob
|
3 |
import json
|
4 |
import os
|
|
|
5 |
from pathlib import Path
|
6 |
|
7 |
import numpy as np
|
|
|
11 |
|
12 |
from models.experimental import attempt_load
|
13 |
from utils.datasets import create_dataloader
|
14 |
+
from utils.general import coco80_to_coco91_class, check_dataset, check_file, check_img_size, compute_loss, \
|
15 |
+
non_max_suppression, scale_coords, xyxy2xywh, clip_coords, plot_images, xywh2xyxy, box_iou, output_to_target, \
|
16 |
+
ap_per_class, set_logging, increment_dir
|
17 |
from utils.torch_utils import select_device, time_synchronized
|
18 |
|
19 |
|
|
|
45 |
device = select_device(opt.device, batch_size=batch_size)
|
46 |
save_txt = opt.save_txt # save *.txt labels
|
47 |
|
48 |
+
# Directories
|
49 |
+
if save_dir == Path('runs/test'): # if default
|
50 |
+
os.makedirs('runs/test', exist_ok=True) # make base
|
51 |
+
save_dir = Path(increment_dir(save_dir / 'exp', opt.name)) # increment run
|
52 |
+
os.makedirs(save_dir / 'labels' if save_txt else save_dir, exist_ok=True) # make new dir
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
# Load model
|
55 |
model = attempt_load(weights, map_location=device) # load FP32 model
|
|
|
138 |
x[:, :4] = scale_coords(img[si].shape[1:], x[:, :4], shapes[si][0], shapes[si][1]) # to original
|
139 |
for *xyxy, conf, cls in x:
|
140 |
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
141 |
+
line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
|
142 |
+
with open(str(save_dir / 'labels' / Path(paths[si]).stem) + '.txt', 'a') as f:
|
143 |
f.write(('%g ' * len(line) + '\n') % line)
|
144 |
|
145 |
# W&B logging
|
|
|
262 |
print('ERROR: pycocotools unable to run: %s' % e)
|
263 |
|
264 |
# Return results
|
265 |
+
print('Results saved to %s' % save_dir)
|
266 |
model.float() # for training
|
267 |
maps = np.zeros(nc) + map
|
268 |
for i, c in enumerate(ap_class):
|
|
|
287 |
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
|
288 |
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
|
289 |
parser.add_argument('--save-dir', type=str, default='runs/test', help='directory to save results')
|
290 |
+
parser.add_argument('--name', default='', help='name to append to --save-dir: i.e. runs/{N} -> runs/{N}_{name}')
|
291 |
opt = parser.parse_args()
|
292 |
opt.save_json |= opt.data.endswith('coco.yaml')
|
293 |
opt.data = check_file(opt.data) # check file
|
|
|
309 |
save_conf=opt.save_conf,
|
310 |
)
|
311 |
|
|
|
|
|
312 |
elif opt.task == 'study': # run over a range of settings and save/plot
|
313 |
for weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']:
|
314 |
f = 'study_%s_%s.txt' % (Path(opt.data).stem, Path(weights).stem) # filename to save to
|
train.py
CHANGED
@@ -1,5 +1,6 @@
|
|
1 |
import argparse
|
2 |
import logging
|
|
|
3 |
import os
|
4 |
import random
|
5 |
import shutil
|
@@ -7,7 +8,6 @@ import time
|
|
7 |
from pathlib import Path
|
8 |
from warnings import warn
|
9 |
|
10 |
-
import math
|
11 |
import numpy as np
|
12 |
import torch.distributed as dist
|
13 |
import torch.nn as nn
|
@@ -404,14 +404,14 @@ if __name__ == '__main__':
|
|
404 |
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
|
405 |
parser.add_argument('--cache-images', action='store_true', help='cache images for faster training')
|
406 |
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
|
407 |
-
parser.add_argument('--name', default='', help='renames experiment folder exp{N} to exp{N}_{name} if supplied')
|
408 |
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
409 |
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
|
410 |
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
|
411 |
parser.add_argument('--adam', action='store_true', help='use torch.optim.Adam() optimizer')
|
412 |
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
|
413 |
parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')
|
414 |
-
parser.add_argument('--logdir', type=str, default='runs/', help='logging directory')
|
|
|
415 |
parser.add_argument('--log-imgs', type=int, default=10, help='number of images for W&B logging, max 100')
|
416 |
parser.add_argument('--workers', type=int, default=8, help='maximum number of dataloader workers')
|
417 |
|
@@ -428,7 +428,7 @@ if __name__ == '__main__':
|
|
428 |
# Resume
|
429 |
if opt.resume: # resume an interrupted run
|
430 |
ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run() # specified or most recent path
|
431 |
-
log_dir = Path(ckpt).parent.parent # runs/exp0
|
432 |
assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
|
433 |
with open(log_dir / 'opt.yaml') as f:
|
434 |
opt = argparse.Namespace(**yaml.load(f, Loader=yaml.FullLoader)) # replace
|
@@ -467,14 +467,13 @@ if __name__ == '__main__':
|
|
467 |
if opt.global_rank in [-1, 0]:
|
468 |
# Tensorboard
|
469 |
logger.info(f'Start Tensorboard with "tensorboard --logdir {opt.logdir}", view at http://localhost:6006/')
|
470 |
-
tb_writer = SummaryWriter(log_dir=log_dir) # runs/exp0
|
471 |
|
472 |
# W&B
|
473 |
try:
|
474 |
import wandb
|
475 |
|
476 |
assert os.environ.get('WANDB_DISABLED') != 'true'
|
477 |
-
logger.info("Weights & Biases logging enabled, to disable set os.environ['WANDB_DISABLED'] = 'true'")
|
478 |
except (ImportError, AssertionError):
|
479 |
opt.log_imgs = 0
|
480 |
logger.info("Install Weights & Biases for experiment logging via 'pip install wandb' (recommended)")
|
|
|
1 |
import argparse
|
2 |
import logging
|
3 |
+
import math
|
4 |
import os
|
5 |
import random
|
6 |
import shutil
|
|
|
8 |
from pathlib import Path
|
9 |
from warnings import warn
|
10 |
|
|
|
11 |
import numpy as np
|
12 |
import torch.distributed as dist
|
13 |
import torch.nn as nn
|
|
|
404 |
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
|
405 |
parser.add_argument('--cache-images', action='store_true', help='cache images for faster training')
|
406 |
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
|
|
|
407 |
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
408 |
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
|
409 |
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
|
410 |
parser.add_argument('--adam', action='store_true', help='use torch.optim.Adam() optimizer')
|
411 |
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
|
412 |
parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')
|
413 |
+
parser.add_argument('--logdir', type=str, default='runs/train', help='logging directory')
|
414 |
+
parser.add_argument('--name', default='', help='name to append to --save-dir: i.e. runs/{N} -> runs/{N}_{name}')
|
415 |
parser.add_argument('--log-imgs', type=int, default=10, help='number of images for W&B logging, max 100')
|
416 |
parser.add_argument('--workers', type=int, default=8, help='maximum number of dataloader workers')
|
417 |
|
|
|
428 |
# Resume
|
429 |
if opt.resume: # resume an interrupted run
|
430 |
ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run() # specified or most recent path
|
431 |
+
log_dir = Path(ckpt).parent.parent # runs/train/exp0
|
432 |
assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
|
433 |
with open(log_dir / 'opt.yaml') as f:
|
434 |
opt = argparse.Namespace(**yaml.load(f, Loader=yaml.FullLoader)) # replace
|
|
|
467 |
if opt.global_rank in [-1, 0]:
|
468 |
# Tensorboard
|
469 |
logger.info(f'Start Tensorboard with "tensorboard --logdir {opt.logdir}", view at http://localhost:6006/')
|
470 |
+
tb_writer = SummaryWriter(log_dir=log_dir) # runs/train/exp0
|
471 |
|
472 |
# W&B
|
473 |
try:
|
474 |
import wandb
|
475 |
|
476 |
assert os.environ.get('WANDB_DISABLED') != 'true'
|
|
|
477 |
except (ImportError, AssertionError):
|
478 |
opt.log_imgs = 0
|
479 |
logger.info("Install Weights & Biases for experiment logging via 'pip install wandb' (recommended)")
|
tutorial.ipynb
CHANGED
@@ -596,22 +596,22 @@
|
|
596 |
}
|
597 |
},
|
598 |
"source": [
|
599 |
-
"!python detect.py --weights yolov5s.pt --img 640 --conf 0.25 --source
|
600 |
-
"Image(filename='
|
601 |
],
|
602 |
"execution_count": null,
|
603 |
"outputs": [
|
604 |
{
|
605 |
"output_type": "stream",
|
606 |
"text": [
|
607 |
-
"Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.25, device='', img_size=640, iou_thres=0.45, save_conf=False, save_dir='
|
608 |
"Using CUDA device0 _CudaDeviceProperties(name='Tesla V100-SXM2-16GB', total_memory=16130MB)\n",
|
609 |
"\n",
|
610 |
"Fusing layers... \n",
|
611 |
"Model Summary: 140 layers, 7.45958e+06 parameters, 0 gradients\n",
|
612 |
-
"image 1/2 /content/yolov5/
|
613 |
-
"image 2/2 /content/yolov5/
|
614 |
-
"Results saved to
|
615 |
"Done. (0.113s)\n"
|
616 |
],
|
617 |
"name": "stdout"
|
@@ -640,7 +640,7 @@
|
|
640 |
"id": "4qbaa3iEcrcE"
|
641 |
},
|
642 |
"source": [
|
643 |
-
"Results are saved to `
|
644 |
"<img src=\"https://user-images.githubusercontent.com/26833433/98274798-2b7a7a80-1f94-11eb-91a4-70c73593e26b.jpg\" width=\"900\"> "
|
645 |
]
|
646 |
},
|
@@ -887,7 +887,7 @@
|
|
887 |
"source": [
|
888 |
"Train a YOLOv5s model on [COCO128](https://www.kaggle.com/ultralytics/coco128) with dataset `--data coco128.yaml`, starting from pretrained `--weights yolov5s.pt`, or from randomly initialized `--weights '' --cfg yolov5s.yaml`. Models are downloaded automatically from the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases), and **COCO, COCO128, and VOC datasets are downloaded automatically** on first use.\n",
|
889 |
"\n",
|
890 |
-
"All training results are saved to `runs/exp0` for the first experiment, then `runs/exp1`, `runs/exp2` etc. for subsequent experiments.\n"
|
891 |
]
|
892 |
},
|
893 |
{
|
@@ -969,7 +969,7 @@
|
|
969 |
"Analyzing anchors... anchors/target = 4.26, Best Possible Recall (BPR) = 0.9946\n",
|
970 |
"Image sizes 640 train, 640 test\n",
|
971 |
"Using 2 dataloader workers\n",
|
972 |
-
"Logging results to runs/exp0\n",
|
973 |
"Starting training for 3 epochs...\n",
|
974 |
"\n",
|
975 |
" Epoch gpu_mem box obj cls total targets img_size\n",
|
@@ -986,8 +986,8 @@
|
|
986 |
" 2/2 3.17G 0.04445 0.06545 0.01666 0.1266 149 640: 100% 8/8 [00:01<00:00, 4.33it/s]\n",
|
987 |
" Class Images Targets P R mAP@.5 mAP@.5:.95: 100% 8/8 [00:02<00:00, 2.78it/s]\n",
|
988 |
" all 128 929 0.395 0.766 0.701 0.455\n",
|
989 |
-
"Optimizer stripped from runs/exp0/weights/last.pt, 15.2MB\n",
|
990 |
-
"Optimizer stripped from runs/exp0/weights/best.pt, 15.2MB\n",
|
991 |
"3 epochs completed in 0.005 hours.\n",
|
992 |
"\n"
|
993 |
],
|
@@ -1030,7 +1030,7 @@
|
|
1030 |
"source": [
|
1031 |
"## Local Logging\n",
|
1032 |
"\n",
|
1033 |
-
"All results are logged by default to the `runs/exp0` directory, with a new directory created for each new training as `runs/exp1`, `runs/exp2`, etc. View train and test jpgs to see mosaics, labels/predictions and augmentation effects. Note a **Mosaic Dataloader** is used for training (shown below), a new concept developed by Ultralytics and first featured in [YOLOv4](https://arxiv.org/abs/2004.10934)."
|
1034 |
]
|
1035 |
},
|
1036 |
{
|
@@ -1039,9 +1039,9 @@
|
|
1039 |
"id": "riPdhraOTCO0"
|
1040 |
},
|
1041 |
"source": [
|
1042 |
-
"Image(filename='runs/exp0/train_batch0.jpg', width=800) # train batch 0 mosaics and labels\n",
|
1043 |
-
"Image(filename='runs/exp0/test_batch0_gt.jpg', width=800) # test batch 0 ground truth\n",
|
1044 |
-
"Image(filename='runs/exp0/test_batch0_pred.jpg', width=800) # test batch 0 predictions"
|
1045 |
],
|
1046 |
"execution_count": null,
|
1047 |
"outputs": []
|
@@ -1078,7 +1078,7 @@
|
|
1078 |
},
|
1079 |
"source": [
|
1080 |
"from utils.utils import plot_results \n",
|
1081 |
-
"plot_results(save_dir='runs/exp0') # plot results.txt as results.png\n",
|
1082 |
"Image(filename='results.png', width=800) "
|
1083 |
],
|
1084 |
"execution_count": null,
|
@@ -1170,9 +1170,9 @@
|
|
1170 |
" for di in 0 cpu # inference devices\n",
|
1171 |
" do\n",
|
1172 |
" python detect.py --weights $x.pt --device $di # detect official\n",
|
1173 |
-
" python detect.py --weights runs/exp0/weights/last.pt --device $di # detect custom\n",
|
1174 |
" python test.py --weights $x.pt --device $di # test official\n",
|
1175 |
-
" python test.py --weights runs/exp0/weights/last.pt --device $di # test custom\n",
|
1176 |
" done\n",
|
1177 |
" python models/yolo.py --cfg $x.yaml # inspect\n",
|
1178 |
" python models/export.py --weights $x.pt --img 640 --batch 1 # export\n",
|
|
|
596 |
}
|
597 |
},
|
598 |
"source": [
|
599 |
+
"!python detect.py --weights yolov5s.pt --img 640 --conf 0.25 --source data/images/\n",
|
600 |
+
"Image(filename='runs/detect/exp0/zidane.jpg', width=600)"
|
601 |
],
|
602 |
"execution_count": null,
|
603 |
"outputs": [
|
604 |
{
|
605 |
"output_type": "stream",
|
606 |
"text": [
|
607 |
+
"Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.25, device='', img_size=640, iou_thres=0.45, save_conf=False, save_dir='runs/detect', save_txt=False, source='data/images/', update=False, view_img=False, weights=['yolov5s.pt'])\n",
|
608 |
"Using CUDA device0 _CudaDeviceProperties(name='Tesla V100-SXM2-16GB', total_memory=16130MB)\n",
|
609 |
"\n",
|
610 |
"Fusing layers... \n",
|
611 |
"Model Summary: 140 layers, 7.45958e+06 parameters, 0 gradients\n",
|
612 |
+
"image 1/2 /content/yolov5/data/images/bus.jpg: 640x480 4 persons, 1 buss, 1 skateboards, Done. (0.012s)\n",
|
613 |
+
"image 2/2 /content/yolov5/data/images/zidane.jpg: 384x640 2 persons, 2 ties, Done. (0.012s)\n",
|
614 |
+
"Results saved to runs/detect/exp0\n",
|
615 |
"Done. (0.113s)\n"
|
616 |
],
|
617 |
"name": "stdout"
|
|
|
640 |
"id": "4qbaa3iEcrcE"
|
641 |
},
|
642 |
"source": [
|
643 |
+
"Results are saved to `runs/detect`. A full list of available inference sources:\n",
|
644 |
"<img src=\"https://user-images.githubusercontent.com/26833433/98274798-2b7a7a80-1f94-11eb-91a4-70c73593e26b.jpg\" width=\"900\"> "
|
645 |
]
|
646 |
},
|
|
|
887 |
"source": [
|
888 |
"Train a YOLOv5s model on [COCO128](https://www.kaggle.com/ultralytics/coco128) with dataset `--data coco128.yaml`, starting from pretrained `--weights yolov5s.pt`, or from randomly initialized `--weights '' --cfg yolov5s.yaml`. Models are downloaded automatically from the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases), and **COCO, COCO128, and VOC datasets are downloaded automatically** on first use.\n",
|
889 |
"\n",
|
890 |
+
"All training results are saved to `runs/train/exp0` for the first experiment, then `runs/exp1`, `runs/exp2` etc. for subsequent experiments.\n"
|
891 |
]
|
892 |
},
|
893 |
{
|
|
|
969 |
"Analyzing anchors... anchors/target = 4.26, Best Possible Recall (BPR) = 0.9946\n",
|
970 |
"Image sizes 640 train, 640 test\n",
|
971 |
"Using 2 dataloader workers\n",
|
972 |
+
"Logging results to runs/train/exp0\n",
|
973 |
"Starting training for 3 epochs...\n",
|
974 |
"\n",
|
975 |
" Epoch gpu_mem box obj cls total targets img_size\n",
|
|
|
986 |
" 2/2 3.17G 0.04445 0.06545 0.01666 0.1266 149 640: 100% 8/8 [00:01<00:00, 4.33it/s]\n",
|
987 |
" Class Images Targets P R mAP@.5 mAP@.5:.95: 100% 8/8 [00:02<00:00, 2.78it/s]\n",
|
988 |
" all 128 929 0.395 0.766 0.701 0.455\n",
|
989 |
+
"Optimizer stripped from runs/train/exp0/weights/last.pt, 15.2MB\n",
|
990 |
+
"Optimizer stripped from runs/train/exp0/weights/best.pt, 15.2MB\n",
|
991 |
"3 epochs completed in 0.005 hours.\n",
|
992 |
"\n"
|
993 |
],
|
|
|
1030 |
"source": [
|
1031 |
"## Local Logging\n",
|
1032 |
"\n",
|
1033 |
+
"All results are logged by default to the `runs/train/exp0` directory, with a new directory created for each new training as `runs/exp1`, `runs/exp2`, etc. View train and test jpgs to see mosaics, labels/predictions and augmentation effects. Note a **Mosaic Dataloader** is used for training (shown below), a new concept developed by Ultralytics and first featured in [YOLOv4](https://arxiv.org/abs/2004.10934)."
|
1034 |
]
|
1035 |
},
|
1036 |
{
|
|
|
1039 |
"id": "riPdhraOTCO0"
|
1040 |
},
|
1041 |
"source": [
|
1042 |
+
"Image(filename='runs/train/exp0/train_batch0.jpg', width=800) # train batch 0 mosaics and labels\n",
|
1043 |
+
"Image(filename='runs/train/exp0/test_batch0_gt.jpg', width=800) # test batch 0 ground truth\n",
|
1044 |
+
"Image(filename='runs/train/exp0/test_batch0_pred.jpg', width=800) # test batch 0 predictions"
|
1045 |
],
|
1046 |
"execution_count": null,
|
1047 |
"outputs": []
|
|
|
1078 |
},
|
1079 |
"source": [
|
1080 |
"from utils.utils import plot_results \n",
|
1081 |
+
"plot_results(save_dir='runs/train/exp0') # plot results.txt as results.png\n",
|
1082 |
"Image(filename='results.png', width=800) "
|
1083 |
],
|
1084 |
"execution_count": null,
|
|
|
1170 |
" for di in 0 cpu # inference devices\n",
|
1171 |
" do\n",
|
1172 |
" python detect.py --weights $x.pt --device $di # detect official\n",
|
1173 |
+
" python detect.py --weights runs/train/exp0/weights/last.pt --device $di # detect custom\n",
|
1174 |
" python test.py --weights $x.pt --device $di # test official\n",
|
1175 |
+
" python test.py --weights runs/train/exp0/weights/last.pt --device $di # test custom\n",
|
1176 |
" done\n",
|
1177 |
" python models/yolo.py --cfg $x.yaml # inspect\n",
|
1178 |
" python models/export.py --weights $x.pt --img 640 --batch 1 # export\n",
|
utils/general.py
CHANGED
@@ -955,9 +955,15 @@ def increment_dir(dir, comment=''):
|
|
955 |
# Increments a directory runs/exp1 --> runs/exp2_comment
|
956 |
n = 0 # number
|
957 |
dir = str(Path(dir)) # os-agnostic
|
|
|
|
|
|
|
|
|
|
|
|
|
958 |
dirs = sorted(glob.glob(dir + '*')) # directories
|
959 |
if dirs:
|
960 |
-
matches = [re.search(r"
|
961 |
idxs = [int(m.groups()[0]) for m in matches if m]
|
962 |
if idxs:
|
963 |
n = max(idxs) + 1 # increment
|
@@ -1262,7 +1268,7 @@ def plot_results_overlay(start=0, stop=0): # from utils.general import *; plot_
|
|
1262 |
|
1263 |
|
1264 |
def plot_results(start=0, stop=0, bucket='', id=(), labels=(), save_dir=''):
|
1265 |
-
# from utils.general import *; plot_results(save_dir='runs/exp0')
|
1266 |
# Plot training 'results*.txt' as seen in https://github.com/ultralytics/yolov5#reproduce-our-training
|
1267 |
fig, ax = plt.subplots(2, 5, figsize=(12, 6))
|
1268 |
ax = ax.ravel()
|
|
|
955 |
# Increments a directory runs/exp1 --> runs/exp2_comment
|
956 |
n = 0 # number
|
957 |
dir = str(Path(dir)) # os-agnostic
|
958 |
+
if os.path.isdir(dir):
|
959 |
+
stem = ''
|
960 |
+
dir += os.sep # removed by Path
|
961 |
+
else:
|
962 |
+
stem = Path(dir).stem
|
963 |
+
|
964 |
dirs = sorted(glob.glob(dir + '*')) # directories
|
965 |
if dirs:
|
966 |
+
matches = [re.search(r"%s(\d+)" % stem, d) for d in dirs]
|
967 |
idxs = [int(m.groups()[0]) for m in matches if m]
|
968 |
if idxs:
|
969 |
n = max(idxs) + 1 # increment
|
|
|
1268 |
|
1269 |
|
1270 |
def plot_results(start=0, stop=0, bucket='', id=(), labels=(), save_dir=''):
|
1271 |
+
# from utils.general import *; plot_results(save_dir='runs/train/exp0')
|
1272 |
# Plot training 'results*.txt' as seen in https://github.com/ultralytics/yolov5#reproduce-our-training
|
1273 |
fig, ax = plt.subplots(2, 5, figsize=(12, 6))
|
1274 |
ax = ax.ravel()
|