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

updates

This commit is contained in:
Glenn Jocher 2019-06-25 12:37:24 +02:00
parent e4cc830690
commit 1d76751e1f
2 changed files with 24 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
train.py
View File

@ -11,19 +11,33 @@ from models import *
from utils.datasets import *
from utils.utils import *
# Hyperparameters: train.py --evolve --epochs 2 --img-size 320, Metrics: 0.204 0.302 0.175 0.234 (square smart)
# Hyperparameters: train.py --data data/coco.data --img-size 320 --single-scale --batch-size 64 --accumulate 1 --epochs 1 --evolve 0.087 0.281 0.109 0.121
hyp = {'giou': .035, # giou loss gain
'xy': 0.20, # xy loss gain
'wh': 0.10, # wh loss gain
'cls': 0.035, # cls loss gain
'cls_pw': 79.0, # cls BCELoss positive_weight
'conf': 1.61, # conf loss gain
'conf_bpw': 3.53, # conf BCELoss positive_weight
'conf_pw': 3.53, # conf BCELoss positive_weight
'iou_t': 0.29, # iou target-anchor training threshold
'lr0': 0.001, # initial learning rate
'lrf': -4., # final learning rate = lr0 * (10 ** lrf)
'momentum': 0.90, # SGD momentum
'weight_decay': 0.0005} # optimizer weight decay
# hyp = {'giou': 1.0, # giou loss gain
# 'xy': 1.0, # xy loss gain
# 'wh': 1.0, # wh loss gain
# 'cls': 1.0, # cls loss gain
# 'cls_pw': 79.0, # cls BCELoss positive_weight
# 'conf': 1.0, # conf loss gain
# 'conf_pw': 6.0, # conf BCELoss positive_weight
# 'iou_t': 0.29, # iou target-anchor training threshold
# 'lr0': 0.001, # initial learning rate
# 'lrf': -4., # final learning rate = lr0 * (10 ** lrf)
# 'momentum': 0.90, # SGD momentum
# 'weight_decay': 0.0005} # optimizer weight decay
def train(
cfg,
@ -329,7 +343,7 @@ if __name__ == '__main__':
# Mutate hyperparameters
old_hyp = hyp.copy()
init_seeds(seed=int(time.time()))
s = [.4, .4, .4, .4, .4, .4, .4, .4 * 0, .4 * 0, .04 * 0, .4 * 0] # fractional sigmas
s = [.4, .4, .4, .4, .4, .4, .4, .4, .4 * 0, .4 * 0, .04 * 0, .4 * 0] # fractional sigmas
for i, k in enumerate(hyp.keys()):
x = (np.random.randn(1) * s[i] + 1) ** 1.1 # plt.hist(x.ravel(), 100)
hyp[k] = hyp[k] * float(x) # vary by about 30% 1sigma

9
utils/utils.py
View File

@ -279,7 +279,8 @@ def compute_loss(p, targets, model, giou_loss=False): # predictions, targets, m
# Define criteria
MSE = nn.MSELoss()
CE = nn.CrossEntropyLoss() # (weight=model.class_weights)
BCE = nn.BCEWithLogitsLoss(pos_weight=ft([h['conf_bpw']]))
BCEcls = nn.BCEWithLogitsLoss(pos_weight=ft([h['cls_pw']]))
BCEconf = nn.BCEWithLogitsLoss(pos_weight=ft([h['conf_pw']]))
# Compute losses
bs = p[0].shape[0] # batch size
@ -301,13 +302,17 @@ def compute_loss(p, targets, model, giou_loss=False): # predictions, targets, m
else:
lxy += (k * h['xy']) * MSE(torch.sigmoid(pi[..., 0:2]), txy[i]) # xy loss
lwh += (k * h['wh']) * MSE(pi[..., 2:4], twh[i]) # wh yolo loss
# tclsm = torch.zeros_like(pi[..., 5:])
# tclsm[range(len(b)), tcls[i]] = 1.0
# lcls += (k * h['cls']) * BCEcls(pi[..., 5:], tclsm) # class_conf loss
lcls += (k * h['cls']) * CE(pi[..., 5:], tcls[i]) # class_conf loss
# # Append to text file
# with open('targets.txt', 'a') as file:
# [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
lconf += (k * h['conf']) * BCE(pi0[..., 4], tconf) # obj_conf loss
lconf += (k * h['conf']) * BCEconf(pi0[..., 4], tconf) # obj_conf loss
loss = lxy + lwh + lconf + lcls
return loss, torch.cat((lxy, lwh, lconf, lcls, loss)).detach()