Bienqq
/
car-detection-bayes
2
1
Fork 0
Code Issues Pull Requests Releases Wiki Activity

updates

This commit is contained in:
Glenn Jocher 2020-01-03 09:19:18 -08:00
parent c0095c2bc9
commit eca1a25dcd
2 changed files with 45 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
train.py
View File

@ -446,7 +446,7 @@ if __name__ == '__main__':
mixed_precision = False mixed_precision = False
# scale hyp['obj'] by img_size (evolved at 320) # scale hyp['obj'] by img_size (evolved at 320)
hyp['obj'] *= opt.img_size / 320. # hyp['obj'] *= opt.img_size / 320.
tb_writer = None tb_writer = None
if not opt.evolve: # Train normally if not opt.evolve: # Train normally

62
utils/utils.py
View File

@ -367,7 +367,7 @@ def compute_loss(p, targets, model): # predictions, targets, model
tcls, tbox, indices, anchor_vec = build_targets(model, targets) tcls, tbox, indices, anchor_vec = build_targets(model, targets)
h = model.hyp # hyperparameters h = model.hyp # hyperparameters
arc = model.arc # # (default, uCE, uBCE) detection architectures arc = model.arc # # (default, uCE, uBCE) detection architectures
red = 'mean' # Loss reduction (sum or mean) red = 'sum' # Loss reduction (sum or mean)
# Define criteria # Define criteria
BCEcls = nn.BCEWithLogitsLoss(pos_weight=ft([h['cls_pw']]), reduction=red) BCEcls = nn.BCEWithLogitsLoss(pos_weight=ft([h['cls_pw']]), reduction=red)
@ -399,7 +399,7 @@ def compute_loss(p, targets, model): # predictions, targets, model
pbox = torch.cat((pxy, pwh), 1) # predicted box pbox = torch.cat((pxy, pwh), 1) # predicted box
giou = bbox_iou(pbox.t(), tbox[i], x1y1x2y2=False, GIoU=True) # giou computation giou = bbox_iou(pbox.t(), tbox[i], x1y1x2y2=False, GIoU=True) # giou computation
lbox += (1.0 - giou).sum() if red == 'sum' else (1.0 - giou).mean() # giou loss lbox += (1.0 - giou).sum() if red == 'sum' else (1.0 - giou).mean() # giou loss
tobj[b, a, gj, gi] = 1.0 # giou.detach().type(tobj.dtype) tobj[b, a, gj, gi] = giou.detach().type(tobj.dtype)
if 'default' in arc and model.nc > 1: # cls loss (only if multiple classes) if 'default' in arc and model.nc > 1: # cls loss (only if multiple classes)
t = torch.zeros_like(ps[:, 5:]) # targets t = torch.zeros_like(ps[:, 5:]) # targets
@ -739,11 +739,28 @@ def coco_single_class_labels(path='../coco/labels/train2014/', label_class=43):
shutil.copyfile(src=img_file, dst='new/images/' + Path(file).name.replace('txt', 'jpg')) # copy images shutil.copyfile(src=img_file, dst='new/images/' + Path(file).name.replace('txt', 'jpg')) # copy images
def kmean_anchors(path='data/coco64.txt', n=9, img_size=(320, 640)): def kmean_anchors(path='../coco/train2017.txt', n=9, img_size=(320, 640)):
# from utils.utils import *; _ = kmean_anchors(n=9) # from utils.utils import *; _ = kmean_anchors()
# Produces a list of target kmeans suitable for use in *.cfg files # Produces a list of target kmeans suitable for use in *.cfg files
from utils.datasets import LoadImagesAndLabels from utils.datasets import LoadImagesAndLabels
from scipy.cluster.vq import kmeans thr = 0.20 # IoU threshold
def print_results(thr, wh, k):
k = k[np.argsort(k.prod(1))] # sort small to large
iou = wh_iou(torch.Tensor(wh), torch.Tensor(k))
max_iou, min_iou = iou.max(1)[0], iou.min(1)[0]
bpr, aat = (max_iou > thr).float().mean(), (iou > thr).float().mean() * n # best possible recall, anch > thr
print('%.2f iou_thr: %.3f best possible recall, %.2f anchors > thr' % (thr, bpr, aat))
print('kmeans anchors (n=%g, img_size=%s, IoU=%.3f/%.3f/%.3f-min/mean/best): ' %
(n, img_size, min_iou.mean(), iou.mean(), max_iou.mean()), end='')
for i, x in enumerate(k):
print('%i,%i' % (round(x[0]), round(x[1])), end=', ' if i < len(k) - 1 else '\n') # use in *.cfg
return k
def fitness(thr, wh, k): # mutation fitness
iou = wh_iou(wh, torch.Tensor(k)).max(1)[0] # max iou
bpr = (iou > thr).float().mean() # best possible recall
return iou.mean() * 0.80 + bpr * 0.20 # weighted combination
# Get label wh # Get label wh
wh = [] wh = []
@ -754,11 +771,18 @@ def kmean_anchors(path='data/coco64.txt', n=9, img_size=(320, 640)):
wh = np.concatenate(wh, 0).repeat(nr, axis=0) # augment 10x wh = np.concatenate(wh, 0).repeat(nr, axis=0) # augment 10x
wh *= np.random.uniform(img_size[0], img_size[1], size=(wh.shape[0], 1)) # normalized to pixels (multi-scale) wh *= np.random.uniform(img_size[0], img_size[1], size=(wh.shape[0], 1)) # normalized to pixels (multi-scale)
# Darknet yolov3.cfg anchors
if n == 9:
k = np.array([[10, 13], [16, 30], [33, 23], [30, 61], [62, 45], [59, 119], [116, 90], [156, 198], [373, 326]])
k = print_results(thr, wh, k)
else:
# Kmeans calculation # Kmeans calculation
from scipy.cluster.vq import kmeans
print('Running kmeans on %g points...' % len(wh)) print('Running kmeans on %g points...' % len(wh))
s = wh.std(0) # sigmas for whitening s = wh.std(0) # sigmas for whitening
k, dist = kmeans(wh / s, n, iter=30) # points, mean distance k, dist = kmeans(wh / s, n, iter=20) # points, mean distance
k = k[np.argsort(k.prod(1))] * s # sort small to large k *= s
k = print_results(thr, wh, k)
# # Plot # # Plot
# k, d = [None] * 20, [None] * 20 # k, d = [None] * 20, [None] * 20
@ -768,21 +792,17 @@ def kmean_anchors(path='data/coco64.txt', n=9, img_size=(320, 640)):
# ax = ax.ravel() # ax = ax.ravel()
# ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.') # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
# Measure IoUs # Evolve
iou = wh_iou(torch.Tensor(wh), torch.Tensor(k)) wh = torch.Tensor(wh)
min_iou, max_iou = iou.min(1)[0], iou.max(1)[0] f, ng = fitness(thr, wh, k), 1000 # fitness, generations
for x in [0.10, 0.20, 0.30]: # iou thresholds for _ in tqdm(range(ng), desc='Evolving anchors'):
print('%.2f iou_thr: %.3f best possible recall, %.2f anchors > thr' % kg = (k.copy() * (1 + np.random.random() * np.random.randn(*k.shape) * 0.20)).clip(min=2.0)
(x, (max_iou > x).float().mean(), (iou > x).float().mean() * n)) # BPR (best possible recall) fg = fitness(thr, wh, kg)
if fg > f:
f, k = fg, kg.copy()
print(fg, list(k.round().reshape(-1)))
k = print_results(thr, wh, k)
# Print
print('kmeans anchors (n=%g, img_size=%s, IoU=%.2f/%.2f/%.2f-min/mean/best): ' %
(n, img_size, min_iou.mean(), iou.mean(), max_iou.mean()), end='')
for i, x in enumerate(k):
print('%i,%i' % (round(x[0]), round(x[1])), end=', ' if i < len(k) - 1 else '\n') # use in *.cfg
# Plot
# plt.hist(biou.numpy().ravel(), 100)
return k return k