from deepvac import AttrDict, new config = new() config.synthesis = AttrDict() #synthesis config.synthesis.total_num = 10 config.synthesis.txt_file = 'your lexicon txt' config.synthesis.images_dir = 'your image path' config.synthesis.video_file = 'your video path' config.synthesis.sample_rate = 1 config.synthesis.fonts_dir = 'your font ttf path' config.synthesis.chars = 'your char set path' config.synthesis.dense_ratio = 0.2 config.synthesis.same_font_ratio = 0.5 config.synthesis.one_way_ratio = 0.3 config.synthesis.border_ratio = 0 config.synthesis.vertical_ratio = 0 config.synthesis.random_space_ratio = 0 config.synthesis.random_space_min = -0.1 config.synthesis.random_space_max = 0.1 config.synthesis.min_font = 8 config.synthesis.max_font = 60
# 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
import torch
import torch.optim as optim
from deepvac import AttrDict, new
from data.dataloader import PseTrainDataset, PseTestDataset
from modules.model_mv3fpn import FpnMobileNetv3
from modules.loss import PSELoss
config = new('PSENetTrain')
## ------------------ common ------------------
config.core.PSENetTrain.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
config.core.PSENetTrain.output_dir = 'output'
config.core.PSENetTrain.log_every = 10
config.core.PSENetTrain.disable_git = True
config.core.PSENetTrain.model_reinterpret_cast = True
config.core.PSENetTrain.cast_state_dict_strict = False
#config.core.PSENetTrain.jit_model_path = "./output/script.pt"
## -------------------- training ------------------
## train runtime
config.core.PSENetTrain.epoch_num = 200
config.core.PSENetTrain.save_num = 1
## -------------------- tensorboard ------------------
# config.core.PSENetTrain.tensorboard_port = "6007"
# config.core.PSENetTrain.tensorboard_ip = None
## -------------------- script and quantize ------------------
config.cast.script_model_dir = "./output/script.pt"
import torch
import torch.optim as optim
from deepvac import AttrDict, new
from data.dataloader import DBTrainDataset, DBTrainCocoDataset, DBTestDataset
from modules.model_db import Resnet18DB, Mobilenetv3LargeDB
from modules.loss import DBLoss
config = new('DBNetTrain')
## ------------------ common ------------------
config.core.DBNetTrain.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
config.core.DBNetTrain.output_dir = 'output'
config.core.DBNetTrain.log_every = 100
config.core.DBNetTrain.disable_git = True
config.core.DBNetTrain.model_reinterpret_cast = True
config.core.DBNetTrain.cast_state_dict_strict = True
# config.core.DBNetTrain.jit_model_path = "./output/script.pt"
## -------------------- training ------------------
## train runtime
config.core.DBNetTrain.epoch_num = 200
config.core.DBNetTrain.save_num = 1
## -------------------- tensorboard ------------------
#config.core.DBNetTrain.tensorboard_port = "6007"
#config.core.DBNetTrain.tensorboard_ip = None
## -------------------- script and quantize ------------------
config.cast.ScriptCast = AttrDict()
import torch
from torchvision import transforms as trans
from deepvac import AttrDict, new
from network import HRNet
config = new('PortraitSegTest')
config.core.PortraitSegTest.disable_git = True
config.core.PortraitSegTest.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
num_classes = 2
config.core.PortraitSegTest.net = HRNet(num_classes=num_classes)
config.core.PortraitSegTest.portrait_mask_output_dir = '<your-portrait-mask-dir>'
config.core.PortraitSegTest.model_path = 'weights/portrait.pth'
config.core.PortraitSegTest.test_loader = ''
config.core.PortraitSegTest.compose = trans.Compose([
trans.Resize([448, 448]),
trans.ToTensor(),
trans.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
config.core.Synthesis = config.core.PortraitSegTest.clone()
config.core.Synthesis.gen_portrait_mask = True
config.core.Synthesis.is_clothes_task = True # True: clothes , False: human
config.core.Synthesis.input_image_dir = '<your-input-image-dir>'
config.core.Synthesis.input_label_dir = '<your-input-label-dir>'
import torch
from deepvac import AttrDict, new
from modules.model import RetinaFaceMobileNet, RetinaFaceResNet
from aug import SynthesisHatComposer
config = new('RetinaTest')
config.core.RetinaTest.disable_git = True
config.core.RetinaTest.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
config.core.RetinaTest.model_path = "/ your face det model path /"
config.core.RetinaTest.confidence_threshold = 0.02
config.core.RetinaTest.nms_threshold = 0.4
config.core.RetinaTest.top_k = 5000
config.core.RetinaTest.keep_top_k = 1
config.core.RetinaTest.max_edge = 2000
config.core.RetinaTest.rgb_means = (104, 117, 123)
config.core.RetinaTest.net = RetinaFaceResNet()
config.core.RetinaTest.test_loader = ''
config.core.Synthesis2D = config.core.RetinaTest.clone()
config.core.Synthesis2D.input_image_dir = '/ your input image path /'
config.core.Synthesis2D.input_hat_mask_dir = '/ your hat mask path /'
config.core.Synthesis2D.input_hat_image_dir = '/ your hat(chartlet) image path /'
config.core.Synthesis2D.output_image_dir = '/ your output image path /'
config.core.Synthesis2D.output_anno_dir = '/ your output annotation path /'
import torch
from deepvac import new, AttrDict
from deepvac.datasets import CocoCVContoursDataset
config = new(None)
config.datasets.CocoCVContoursDataset = AttrDict()
config.datasets.CocoCVContoursDataset.auto_detect_subdir_with_basenum = 0
sample_path_prefix_list = ['your sample path prefix list']
target_path_list = ['your json file path list']
config.output_label_dir = 'your output label dir'
config.output_image_dir = 'your output image dir'
config.show = True
config.test_loader_list = []
for i in range(len(sample_path_prefix_list)):
test_dataset = CocoCVContoursDataset(config, sample_path_prefix_list[i], target_path_list[i])
config.test_loader_list.append(torch.utils.data.DataLoader(test_dataset, batch_size=1, pin_memory=False))