339 lines
15 KiB
Python
339 lines
15 KiB
Python
import argparse
|
|
import yaml
|
|
import math
|
|
import os
|
|
import struct
|
|
import torch
|
|
from utils.torch_utils import select_device
|
|
|
|
|
|
def parse_args():
|
|
parser = argparse.ArgumentParser(description="PyTorch YOLOv5 conversion")
|
|
parser.add_argument("-w", "--weights", required=True, help="Input weights (.pt) file path (required)")
|
|
parser.add_argument("-c", "--yaml", help="Input cfg (.yaml) file path")
|
|
parser.add_argument("-mw", "--width", help="Model width (default = 640 / 1280 [P6])")
|
|
parser.add_argument("-mh", "--height", help="Model height (default = 640 / 1280 [P6])")
|
|
parser.add_argument("-mc", "--channels", help="Model channels (default = 3)")
|
|
parser.add_argument("--p6", action="store_true", help="P6 model")
|
|
args = parser.parse_args()
|
|
if not os.path.isfile(args.weights):
|
|
raise SystemExit("Invalid weights file")
|
|
if not args.yaml:
|
|
args.yaml = ""
|
|
if not args.width:
|
|
args.width = 1280 if args.p6 else 640
|
|
if not args.height:
|
|
args.height = 1280 if args.p6 else 640
|
|
if not args.channels:
|
|
args.channels = 3
|
|
return args.weights, args.yaml, args.width, args.height, args.channels, args.p6
|
|
|
|
|
|
def get_width(x, gw, divisor=8):
|
|
return int(math.ceil((x * gw) / divisor)) * divisor
|
|
|
|
|
|
def get_depth(x, gd):
|
|
if x == 1:
|
|
return 1
|
|
r = int(round(x * gd))
|
|
if x * gd - int(x * gd) == 0.5 and int(x * gd) % 2 == 0:
|
|
r -= 1
|
|
return max(r, 1)
|
|
|
|
|
|
pt_file, yaml_file, model_width, model_height, model_channels, p6 = parse_args()
|
|
|
|
model_name = pt_file.split(".pt")[0]
|
|
wts_file = model_name + ".wts"
|
|
cfg_file = model_name + ".cfg"
|
|
|
|
if yaml_file == "":
|
|
yaml_file = "models/" + model_name + ".yaml"
|
|
if not os.path.isfile(yaml_file):
|
|
yaml_file = "models/hub/" + model_name + ".yaml"
|
|
if not os.path.isfile(yaml_file):
|
|
raise SystemExit("YAML file not found")
|
|
elif not os.path.isfile(yaml_file):
|
|
raise SystemExit("Invalid YAML file")
|
|
|
|
device = select_device("cpu")
|
|
model = torch.load(pt_file, map_location=device)["model"].float()
|
|
model.to(device).eval()
|
|
|
|
with open(wts_file, "w") as f:
|
|
wts_write = ""
|
|
conv_count = 0
|
|
cv1 = ""
|
|
cv3 = ""
|
|
cv3_idx = 0
|
|
sppf_idx = 11 if p6 else 9
|
|
for k, v in model.state_dict().items():
|
|
if not "num_batches_tracked" in k and not "anchors" in k and not "anchor_grid" in k:
|
|
vr = v.reshape(-1).cpu().numpy()
|
|
idx = int(k.split(".")[1])
|
|
if ".cv1." in k and not ".m." in k and idx != sppf_idx:
|
|
cv1 += "{} {} ".format(k, len(vr))
|
|
for vv in vr:
|
|
cv1 += " "
|
|
cv1 += struct.pack(">f", float(vv)).hex()
|
|
cv1 += "\n"
|
|
conv_count += 1
|
|
elif cv1 != "" and ".m." in k:
|
|
wts_write += cv1
|
|
cv1 = ""
|
|
if ".cv3." in k:
|
|
cv3 += "{} {} ".format(k, len(vr))
|
|
for vv in vr:
|
|
cv3 += " "
|
|
cv3 += struct.pack(">f", float(vv)).hex()
|
|
cv3 += "\n"
|
|
cv3_idx = idx
|
|
conv_count += 1
|
|
elif cv3 != "" and cv3_idx != idx:
|
|
wts_write += cv3
|
|
cv3 = ""
|
|
cv3_idx = 0
|
|
if not ".cv3." in k and not (".cv1." in k and not ".m." in k and idx != sppf_idx):
|
|
wts_write += "{} {} ".format(k, len(vr))
|
|
for vv in vr:
|
|
wts_write += " "
|
|
wts_write += struct.pack(">f", float(vv)).hex()
|
|
wts_write += "\n"
|
|
conv_count += 1
|
|
f.write("{}\n".format(conv_count))
|
|
f.write(wts_write)
|
|
|
|
with open(cfg_file, "w") as c:
|
|
with open(yaml_file, "r") as f:
|
|
nc = 0
|
|
depth_multiple = 0
|
|
width_multiple = 0
|
|
anchors = ""
|
|
masks = []
|
|
num = 0
|
|
detections = []
|
|
layers = []
|
|
f = yaml.load(f,Loader=yaml.FullLoader)
|
|
c.write("[net]\n")
|
|
c.write("width=%d\n" % model_width)
|
|
c.write("height=%d\n" % model_height)
|
|
c.write("channels=%d\n" % model_channels)
|
|
for l in f:
|
|
if l == "nc":
|
|
nc = f[l]
|
|
elif l == "depth_multiple":
|
|
depth_multiple = f[l]
|
|
elif l == "width_multiple":
|
|
width_multiple = f[l]
|
|
elif l == "anchors":
|
|
a = []
|
|
for v in f[l]:
|
|
a.extend(v)
|
|
mask = []
|
|
for _ in range(int(len(v) / 2)):
|
|
mask.append(num)
|
|
num += 1
|
|
masks.append(mask)
|
|
anchors = str(a)[1:-1]
|
|
elif l == "backbone" or l == "head":
|
|
for v in f[l]:
|
|
if v[2] == "Conv":
|
|
layer = ""
|
|
blocks = 0
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0], width_multiple)
|
|
layer += "size=%d\n" % v[3][1]
|
|
layer += "stride=%d\n" % v[3][2]
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
layers.append([layer, blocks])
|
|
elif v[2] == "C3":
|
|
layer = ""
|
|
blocks = 0
|
|
layer += "\n# C3\n"
|
|
# SPLIT
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0] / 2, width_multiple)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
layer += "\n[route]\n"
|
|
layer += "layers=-2\n"
|
|
blocks += 1
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0] / 2, width_multiple)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
# Residual Block
|
|
if len(v[3]) == 1 or v[3][1] == True:
|
|
for _ in range(get_depth(v[1], depth_multiple)):
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0] / 2, width_multiple)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0] / 2, width_multiple)
|
|
layer += "size=3\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
layer += "\n[shortcut]\n"
|
|
layer += "from=-3\n"
|
|
layer += "activation=linear\n"
|
|
blocks += 1
|
|
# Merge
|
|
layer += "\n[route]\n"
|
|
layer += "layers=-1, -%d\n" % (3 * get_depth(v[1], depth_multiple) + 3)
|
|
blocks += 1
|
|
else:
|
|
for _ in range(get_depth(v[1], depth_multiple)):
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0] / 2, width_multiple)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0] / 2, width_multiple)
|
|
layer += "size=3\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
# Merge
|
|
layer += "\n[route]\n"
|
|
layer += "layers=-1, -%d\n" % (2 * get_depth(v[1], depth_multiple) + 3)
|
|
blocks += 1
|
|
# Transition
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0], width_multiple)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
layer += "\n##########\n"
|
|
blocks += 1
|
|
layers.append([layer, blocks])
|
|
elif v[2] == "SPPF":
|
|
layer = ""
|
|
blocks = 0
|
|
layer += "\n# SPPF\n"
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % (get_width(v[3][0], width_multiple) / 2)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
blocks += 1
|
|
layer += "\n[maxpool]\n"
|
|
layer += "stride=1\n"
|
|
layer += "size=%d\n" % v[3][1]
|
|
blocks += 1
|
|
layer += "\n[maxpool]\n"
|
|
layer += "stride=1\n"
|
|
layer += "size=%d\n" % v[3][1]
|
|
blocks += 1
|
|
layer += "\n[maxpool]\n"
|
|
layer += "stride=1\n"
|
|
layer += "size=%d\n" % v[3][1]
|
|
blocks += 1
|
|
layer += "\n[route]\n"
|
|
layer += "layers=-4, -3, -2, -1\n"
|
|
blocks += 1
|
|
layer += "\n[convolutional]\n"
|
|
layer += "batch_normalize=1\n"
|
|
layer += "filters=%d\n" % get_width(v[3][0], width_multiple)
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "activation=silu\n"
|
|
layer += "\n##########\n"
|
|
blocks += 1
|
|
layers.append([layer, blocks])
|
|
elif v[2] == "nn.Upsample":
|
|
layer = ""
|
|
blocks = 0
|
|
layer += "\n[upsample]\n"
|
|
layer += "stride=%d\n" % v[3][1]
|
|
blocks += 1
|
|
layers.append([layer, blocks])
|
|
elif v[2] == "Concat":
|
|
layer = ""
|
|
blocks = 0
|
|
route = v[0][1]
|
|
r = 0
|
|
if route > 0:
|
|
for i, item in enumerate(layers):
|
|
if i <= route:
|
|
r += item[1]
|
|
else:
|
|
break
|
|
else:
|
|
route = len(layers) + route
|
|
for i, item in enumerate(layers):
|
|
if i <= route:
|
|
r += item[1]
|
|
else:
|
|
break
|
|
layer += "\n# Concat\n"
|
|
layer += "\n[route]\n"
|
|
layer += "layers=-1, %d\n" % (r - 1)
|
|
layer += "\n##########\n"
|
|
blocks += 1
|
|
layers.append([layer, blocks])
|
|
elif v[2] == "Detect":
|
|
for i, n in enumerate(v[0]):
|
|
layer = ""
|
|
blocks = 0
|
|
r = 0
|
|
for j, item in enumerate(layers):
|
|
if j <= n:
|
|
r += item[1]
|
|
else:
|
|
break
|
|
layer += "\n# Detect\n"
|
|
layer += "\n[route]\n"
|
|
layer += "layers=%d\n" % (r - 1)
|
|
blocks += 1
|
|
layer += "\n[convolutional]\n"
|
|
layer += "size=1\n"
|
|
layer += "stride=1\n"
|
|
layer += "pad=1\n"
|
|
layer += "filters=%d\n" % ((nc + 5) * 3)
|
|
layer += "activation=logistic\n"
|
|
blocks += 1
|
|
layer += "\n[yolo]\n"
|
|
layer += "mask=%s\n" % str(masks[i])[1:-1]
|
|
layer += "anchors=%s\n" % anchors
|
|
layer += "classes=%d\n" % nc
|
|
layer += "num=%d\n" % num
|
|
layer += "scale_x_y=2.0\n"
|
|
layer += "beta_nms=0.6\n"
|
|
layer += "new_coords=1\n"
|
|
layer += "\n##########\n"
|
|
blocks += 1
|
|
layers.append([layer, blocks])
|
|
for layer in layers:
|
|
c.write(layer[0])
|