-
Notifications
You must be signed in to change notification settings - Fork 0
/
infer_yolor_process.py
243 lines (208 loc) · 9.41 KB
/
infer_yolor_process.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
# Copyright (C) 2021 Ikomia SAS
# Contact: https://www.ikomia.com
#
# This file is part of the IkomiaStudio software.
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
from ikomia import utils, core, dataprocess
import copy
import torch
from pathlib import Path
import os
#from infer_yolor.yolor.utils.google_utils import gdrive_download
from infer_yolor.yolor.models.models import *
from infer_yolor.yolor.utils.torch_utils import select_device
import numpy as np
from torchvision.transforms import Resize
from infer_yolor.yolor.utils.general import non_max_suppression, scale_coords
import requests
# --------------------
# - Class to handle the process parameters
# - Inherits PyCore.CWorkflowTaskParam from Ikomia API
# --------------------
class YoloRParam(core.CWorkflowTaskParam):
def __init__(self):
core.CWorkflowTaskParam.__init__(self)
# Place default value initialization here
self.update = False
self.config_file = ""
self.model_weight_file = ""
self.dataset = "COCO"
self.input_size = 512
self.conf_thres = 0.25
self.iou_thres = 0.45
self.agnostic_nms = False
self.cuda = torch.cuda.is_available()
def set_values(self, params):
# Set parameters values from Ikomia application
# Parameters values are stored as string and accessible like a python dict
self.input_size = int(params["input_size"])
self.config_file = str(params["config_file"])
self.model_weight_file = str(params["model_weight_file"])
self.dataset = str(params["dataset"])
self.conf_thres = float(params["conf_thres"])
self.iou_thres = float(params["iou_thresh"])
self.agnostic_nms = utils.strtobool(params["agnostic_nms"])
self.cuda = utils.strtobool(params["cuda"])
def get_values(self):
# Send parameters values to Ikomia application
# Create the specific dict structure (string container)
params = {
"input_size": str(self.input_size),
"config_file": self.config_file,
"model_weight_file": self.model_weight_file,
"dataset": self.dataset,
"conf_thres": str(self.conf_thres),
"iou_thresh": str(self.iou_thres),
"agnostic_nms": str(self.agnostic_nms),
"cuda": str(self.cuda)
}
return params
# --------------------
# - Class which implements the process
# - Inherits PyCore.CWorkflowTask or derived from Ikomia API
# --------------------
class YoloRProcess(dataprocess.CObjectDetectionTask):
def __init__(self, name, param):
dataprocess.CObjectDetectionTask.__init__(self, name)
self.model = None
self.update = False
self.config_file = None
self.model_weight_file = ""
self.model_path = None
# Detect if we have a GPU available
self.device = torch.device("cpu")
# Create parameters class
if param is None:
self.set_param_object(YoloRParam())
else:
self.set_param_object(copy.deepcopy(param))
def get_progress_steps(self):
# Function returning the number of progress steps for this process
# This is handled by the main progress bar of Ikomia application
return 2
def run(self):
# Core function of your process
# Call beginTaskRun for initialization
self.begin_task_run()
# Temporary fix to clean detection outputs
self.get_output(1).clear_data()
# Get parameters :
param = self.get_param_object()
# Display all classes
classes = None
if param.model_weight_file != "":
param.dataset ="Custom"
if param.dataset == "COCO":
# Get weight_path
self.model_path = Path(
os.path.dirname(os.path.realpath(__file__)), "yolor", "models", "yolor_p6.pt")
if not os.path.isfile(self.model_path):
model_url = utils.get_model_hub_url() + "/" + self.name + "/yolor_p6.pt"
print("Downloading weights...")
response = requests.get(model_url, stream=True)
with open(self.model_path, 'wb') as f:
for chunk in response.iter_content(chunk_size=8192):
f.write(chunk)
print("Weights downloaded")
self.config_file = Path(os.path.dirname(os.path.realpath(__file__)) + "/yolor/cfg/" + "yolor_p6.cfg")
name_file_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "yolor", "data", "coco.names")
self.read_class_names(name_file_path)
ckpt = torch.load(self.model_path)
if param.dataset == "Custom":
self.config_file = param.config_file
self.model_path = param.model_weight_file
ckpt = torch.load(self.model_path)
if 'names' in ckpt.keys():
self.set_names(ckpt['names'])
if self.model is None or param.update:
self.device = torch.device("cuda") if param.cuda else torch.device("cpu")
self.model = Darknet(self.config_file.__str__()).to(self.device)
self.model.eval()
# state_dict = {k: v for k, v in ckpt['model'].items() if self.model.state_dict()[k].numel() == v.numel()}
state_dict = ckpt['model']
self.model.load_state_dict(state_dict, strict=True)
print('Transferred %g/%g items from %s' %
(len(state_dict), len(self.model.state_dict()), self.model_path)) # report
param.update = False
self.emit_step_progress()
if self.model is not None:
img_input = self.get_input(0)
src_image = img_input.get_image()
self.model.to(self.device)
with torch.no_grad():
self.detect(src_image, param.input_size, param.conf_thres, param.iou_thres, classes, param.agnostic_nms)
# Call endTaskRun to finalize process
self.emit_step_progress()
self.end_task_run()
def detect(self, im0, imgsz, conf_thres, iou_thres, classes, agnostic_nms):
half = False # for this model half precision does not work in pytorch 1.9
if half:
self.model.half() # to FP16
# Run inference
h, w, _ = np.shape(im0)
img = np.ascontiguousarray(im0)
img = torch.from_numpy(img)
img = img.to(self.device)
img = img.half() if half else img.float() # uint8 to fp32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
img = img.permute(0, 3, 1, 2)
img = Resize((imgsz, imgsz))(img)
inf_out = self.model(img)[0].to('cpu')
output = non_max_suppression(inf_out, conf_thres=conf_thres, iou_thres=iou_thres, classes=classes,
agnostic=agnostic_nms)[0]
# Rescale boxes from img_size to im0 size
whwh = torch.tensor([w / imgsz, h / imgsz, w / imgsz, h / imgsz]).to('cpu')
for pred in output:
pred[:4] *= whwh
index = 0
for *xyxy, conf, cls in output:
# Box
w = float(xyxy[2] - xyxy[0])
h = float(xyxy[3] - xyxy[1])
self.add_object(index, int(cls), conf.item(), float(xyxy[0]), float(xyxy[1]), w, h)
index += 1
# --------------------
# - Factory class to build process object
# - Inherits PyDataProcess.CTaskFactory from Ikomia API
# --------------------
class YoloRProcessFactory(dataprocess.CTaskFactory):
def __init__(self):
dataprocess.CTaskFactory.__init__(self)
# Set process information as string here
self.info.name = "infer_yolor"
self.info.short_description = "Inference for YoloR object detection models"
self.info.authors = "Chien-Yao Wang, I-Hau Yeh, Hong-Yuan Mark Liao"
# relative path -> as displayed in Ikomia application process tree
self.info.path = "Plugins/Python/Detection"
self.info.version = "1.1.7"
self.info.icon_path = "icons/icon.png"
self.info.article = "You Only Learn One Representation: Unified Network for Multiple Tasks"
self.info.journal = "Arxiv"
self.info.year = 2021
self.info.license = "GPL-3.0 License"
# URL of documentation
self.info.documentation_link = "https://arxiv.org/abs/2105.04206"
# Code source repository
self.info.repository = "https://github.com/Ikomia-hub/infer_yolor"
self.info.original_repository = "https://github.com/WongKinYiu/yolor"
# Keywords used for search
self.info.keywords = "yolo, inference, pytorch, object, detection"
self.info.algo_type = core.AlgoType.INFER
self.info.algo_tasks = "OBJECT_DETECTION"
def create(self, param=None):
# Create process object
return YoloRProcess(self.info.name, param)