-
Notifications
You must be signed in to change notification settings - Fork 2
/
config.py
167 lines (144 loc) · 7.96 KB
/
config.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
#! /usr/bin/python3
# -*- coding:utf-8 -*-
# import first: system libs
import math
# import second: third party libs
import torch
from torchvision import transforms
# import third: libs in your program
from deepvac import AttrDict, new
from deepvac.aug.yolo_aug import *
from aug.aug import Yolov5TrainComposer, Yolov5ValComposer
from data.datasets import Yolov5MosaicDataset, Yolov5Dataset
from modules import Yolov5S, Yolov5L, Yolov5Loss
################################################################################
### TRAIN
################################################################################
config = new("Yolov5Train")
### ---------------------------------- common ----------------------------------
config.core.Yolov5Train.class_num = 80
config.core.Yolov5Train.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
config.core.Yolov5Train.output_dir = "output"
config.core.Yolov5Train.log_every = 10
config.core.Yolov5Train.disable_git = True
config.core.Yolov5Train.model_reinterpret_cast = True
config.core.Yolov5Train.cast_state_dict_strict = True
### ---------------------------------- training --------------------------------
config.core.Yolov5Train.ema = True
# define ema_decay with other func
# config.ema_decay = lambda x: 0.9999 * (1 - math.exp(-x / 2000))
config.core.Yolov5Train.amp = False
config.core.Yolov5Train.epoch_num = 150
config.core.Yolov5Train.save_num = 1
config.core.Yolov5Train.shuffle = True
config.core.Yolov5Train.batch_size = 16
config.core.Yolov5Train.num_workers = 8
config.core.Yolov5Train.pin_memory = True
config.core.Yolov5Train.nominal_batch_factor = 4
config.core.Yolov5Train.model_path = "output/trained.pth"
# load script and quantize model path
# config.core.Yolov5Train.jit_model_path = "<pretrained-model-path>"
### ---------------------------------- tensorboard -----------------------------
# config.core.Yolov5Train.tensorboard_port = "6007"
# config.core.Yolov5Train.tensorboard_ip = None
### ------------------------------ script and quantize -------------------------
# config.cast.TraceCast = AttrDict()
# config.cast.TraceCast.model_dir = "./script.pt"
# config.cast.TraceCast.static_quantize_dir = "./script.sq"
# config.cast.TraceCast.dynamic_quantize_dir = "./quantize.sq"
### ---------------------------------- dataset ---------------------------------
config.core.Yolov5Train.img_size = 416
config.core.Yolov5Train.border = [-config.core.Yolov5Train.img_size / 2] * 2
config.aug.HSVAug = AttrDict()
config.aug.HSVAug.hgain = 0.015
config.aug.HSVAug.sgain = 0.7
config.aug.HSVAug.vgain = 0.4
config.aug.YoloPerspectiveAug = AttrDict()
config.aug.YoloPerspectiveAug.scale = 0.5
config.aug.YoloPerspectiveAug.shear = 0.0
config.aug.YoloPerspectiveAug.degrees = 0.0
config.aug.YoloPerspectiveAug.translate = 0.1
config.aug.YoloPerspectiveAug.perspective = 0.0
config.aug.YoloPerspectiveAug.border = config.core.Yolov5Train.border
config.datasets.Yolov5MosaicDataset = AttrDict()
config.datasets.Yolov5MosaicDataset.composer = Yolov5TrainComposer(config)
train_sample_path = "data/coco/images/train2017"
train_target_path = "data/coco/instances_train2017.json"
config.core.Yolov5Train.train_dataset = Yolov5MosaicDataset(config, train_sample_path, train_target_path, config.core.Yolov5Train.img_size, config.core.Yolov5Train.border)
config.core.Yolov5Train.train_loader = torch.utils.data.DataLoader(config.core.Yolov5Train.train_dataset,
batch_size=config.core.Yolov5Train.batch_size,
shuffle=config.core.Yolov5Train.shuffle,
num_workers=config.core.Yolov5Train.num_workers,
pin_memory=config.core.Yolov5Train.pin_memory,
collate_fn=Yolov5MosaicDataset.collate_fn)
config.datasets.Yolov5Dataset = AttrDict()
config.datasets.Yolov5Dataset.composer = Yolov5ValComposer(config)
val_sample_path = "data/coco/images/val2017"
val_target_path = "data/coco/instances_val2017.json"
config.core.Yolov5Train.val_dataset = Yolov5Dataset(config, val_sample_path, val_target_path, config.core.Yolov5Train.img_size)
config.core.Yolov5Train.val_loader = torch.utils.data.DataLoader(config.core.Yolov5Train.val_dataset,
batch_size=config.core.Yolov5Train.batch_size,
num_workers=config.core.Yolov5Train.num_workers,
collate_fn=Yolov5Dataset.collate_fn)
### ---------------------------------- model -----------------------------------
config.core.Yolov5Train.strides = [8, 16, 32]
# support model include (Yolov5S, Yolov5L) now
config.core.Yolov5Train.net = Yolov5S(config.core.Yolov5Train.class_num, config.core.Yolov5Train.strides)
### ---------------------------------- optimizer -------------------------------
config.core.Yolov5Train.lr = 0.01
config.core.Yolov5Train.momentum = 0.937
config.core.Yolov5Train.nesterov = True
config.core.Yolov5Train.optimizer = torch.optim.SGD(config.core.Yolov5Train.net.pg0, lr=config.core.Yolov5Train.lr, momentum=config.core.Yolov5Train.momentum, nesterov=config.core.Yolov5Train.nesterov)
config.core.Yolov5Train.optimizer.add_param_group({'params': config.core.Yolov5Train.net.pg1, 'weight_decay': 5e-4})
config.core.Yolov5Train.optimizer.add_param_group({'params': config.core.Yolov5Train.net.pg2})
del config.core.Yolov5Train.net.pg0
del config.core.Yolov5Train.net.pg1
del config.core.Yolov5Train.net.pg2
### ---------------------------------- scheduler -------------------------------
config.core.Yolov5Train.lr_lambda = lambda x: ((1 + math.cos(x * math.pi / config.core.Yolov5Train.epoch_num)) / 2) * (1 - 0.2) + 0.2
config.core.Yolov5Train.scheduler = torch.optim.lr_scheduler.LambdaLR(config.core.Yolov5Train.optimizer, lr_lambda=config.core.Yolov5Train.lr_lambda)
config.core.Yolov5Train.warmup_bias_lr = 0.1
config.core.Yolov5Train.warmup_momentum = 0.8
config.core.Yolov5Train.warmup_iter = max(3 * len(config.core.Yolov5Train.train_loader), 1e3)
### ---------------------------------- loss ------------------------------------
cls_scale = 0.5 * config.core.Yolov5Train.class_num / 80
box_scale = 0.05
obj_scale = 1.0
config.core.Yolov5Train.criterion = Yolov5Loss(config, config.core.Yolov5Train.net.detect, cls_scale, box_scale, obj_scale, config.core.Yolov5Train.strides, config.core.Yolov5Train.device)
################################################################################
### TEST
################################################################################
config.core.Yolov5Test = AttrDict()
### ---------------------------------- test ------------------------------------
config.core.Yolov5Test.device = "cpu"
config.core.Yolov5Test.class_num = 80
config.core.Yolov5Test.img_size = 640
config.core.Yolov5Test.strides = [8, 16, 32]
config.core.Yolov5Test.model_reinterpret_cast = True
config.core.Yolov5Test.cast_state_dict_strict = True
config.core.Yolov5Test.net = Yolov5S(config.core.Yolov5Test.class_num, config.core.Yolov5Test.strides)
config.core.Yolov5Test.model_path = "output/trained.pth"
config.core.Yolov5Test.test_sample_path = "your test sample path"
config.core.Yolov5Test.half = False
config.core.Yolov5Test.show_output_dir = "output/show"
config.core.Yolov5Test.iou_thres = 0.45
config.core.Yolov5Test.conf_thres = 0.25
config.core.Yolov5Test.idx2cat = ["cls{}".format(i) for i in range(config.core.Yolov5Test.class_num)]
################################################################################
### CAST
config.core.Yolov5Train.cast2cpu = True
config.core.Yolov5Test.cast2cpu = True
################################################################################
import coremltools
config.cast.CoremlCast = AttrDict()
config.cast.TraceCast = AttrDict()
config.cast.TraceCast.model_dir = "output/trace.pt"
config.cast.CoremlCast.model_dir = "output/coreml.mlmodel"
config.cast.CoremlCast.input_type = None
config.cast.CoremlCast.scale = 1.0 / 255.0
config.cast.CoremlCast.color_layout = 'BGR'
config.cast.CoremlCast.blue_bias = 0
config.cast.CoremlCast.green_bias = 0
config.cast.CoremlCast.red_bias = 0
config.cast.CoremlCast.minimum_deployment_target = coremltools.target.iOS13
config.cast.CoremlCast.classfier_config = ["cls{}".format(i) for i in range(config.core.Yolov5Test.class_num)]