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

updates

This commit is contained in:
Glenn Jocher 2020-03-09 16:00:05 -07:00
parent 67e7ac221f
commit 6130b70fe7
1 changed files with 5 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
models.py
View File

@ -87,24 +87,13 @@ def create_modules(module_defs, img_size, arc):
# Initialize preceding Conv2d() bias (https://arxiv.org/pdf/1708.02002.pdf section 3.3) # Initialize preceding Conv2d() bias (https://arxiv.org/pdf/1708.02002.pdf section 3.3)
try: try:
p = math.log(1 / (modules.nc - 0.99)) # class probability -> sigmoid(p) = 1/nc bo = -4.5 #  obj bias
if arc == 'default' or arc == 'Fdefault': # default bc = math.log(1 / (modules.nc - 0.99)) # cls bias: class probability is sigmoid(p) = 1/nc
b = [-4.5, p] # obj, cls
elif arc == 'uBCE': # unified BCE (80 classes)
b = [0, -9.0]
elif arc == 'uCE': # unified CE (1 background + 80 classes)
b = [10, -0.1]
elif arc == 'uFBCE': # unified FocalBCE (5120 obj, 80 classes)
b = [0, -6.5]
elif arc == 'uFCE': # unified FocalCE (64 cls, 1 background + 80 classes)
b = [7.7, -1.1]
bias = module_list[-1][0].bias.view(len(mask), -1) # 255 to 3x85 bias = module_list[-1][0].bias.view(len(mask), -1) # 255 to 3x85
bias[:, 4] += b[0] - bias[:, 4].mean() # obj bias[:, 4] += bo - bias[:, 4].mean() # obj
bias[:, 5:] += b[1] - bias[:, 5:].mean() # cls bias[:, 5:] += bc - bias[:, 5:].mean() # cls
# bias = torch.load('weights/yolov3-spp.bias.pt')[yolo_index] # list of tensors [3x85, 3x85, 3x85] module_list[-1][0].bias = torch.nn.Parameter(bias.view(-1)) # utils.print_model_biases(model)
module_list[-1][0].bias = torch.nn.Parameter(bias.view(-1))
# utils.print_model_biases(model)
except: except:
print('WARNING: smart bias initialization failure.') print('WARNING: smart bias initialization failure.')