lichen/ieemoo-ai-isempty
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update

Browse Source
This commit is contained in:
李晨
2023-09-21 18:28:43 +08:00
parent 6bbbb07fde
commit 92b8b5098e
6 changed files with 229 additions and 252 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
predict.py
View File

@ -9,24 +9,22 @@ from sklearn.metrics import f1_score
from PIL import Image
from torchvision import transforms
from models.modeling import VisionTransformer, CONFIGS
import lightrise
#模型预测
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--img_size", default=600, type=int, help="Resolution size")
parser.add_argument("--img_size", default=448, type=int, help="Resolution size")
parser.add_argument('--split', type=str, default='overlap', help="Split method") # non-overlap
parser.add_argument('--slide_step', type=int, default=2, help="Slide step for overlap split")
parser.add_argument('--slide_step', type=int, default=12, help="Slide step for overlap split")
parser.add_argument('--smoothing_value', type=float, default=0.0, help="Label smoothing value\n")
parser.add_argument("--pretrained_model", type=str, default="../module/ieemoo-ai-isempty/model/new/ieemooempty_vit_checkpoint.pth", help="load pretrained model")
#parser.add_argument("--pretrained_model", type=str, default="output/ieemooempty_vit_checkpoint.pth", help="load pretrained model") #使用自定义VIT
parser.add_argument("--pretrained_model", type=str, default="output/emptyjudge5_checkpoint.bin", help="load pretrained model")
return parser.parse_args()
class Predictor(object):
def __init__(self, args):
self.args = args
self.args.device = torch.device("cuda")
self.args.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print("self.args.device =", self.args.device)
self.args.nprocs = torch.cuda.device_count()
@ -34,7 +32,7 @@ class Predictor(object):
self.num_classes = 0
self.model = None
self.prepare_model()
self.test_transform = transforms.Compose([transforms.Resize((600, 600), Image.BILINEAR),
self.test_transform = transforms.Compose([transforms.Resize((448, 448), Image.BILINEAR),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
@ -42,14 +40,28 @@ class Predictor(object):
config = CONFIGS["ViT-B_16"]
config.split = self.args.split
config.slide_step = self.args.slide_step
self.num_classes = 5
self.cls_dict = {0: "noemp", 1: "yesemp"}
model_name = os.path.basename(self.args.pretrained_model).replace("_checkpoint.bin", "")
print("use model_name: ", model_name)
if model_name.lower() == "emptyJudge5".lower():
self.num_classes = 5
self.cls_dict = {0: "noemp", 1: "yesemp", 2: "hard", 3: "fly", 4: "stack"}
elif model_name.lower() == "emptyJudge4".lower():
self.num_classes = 4
self.cls_dict = {0: "noemp", 1: "yesemp", 2: "hard", 3: "stack"}
elif model_name.lower() == "emptyJudge3".lower():
self.num_classes = 3
self.cls_dict = {0: "noemp", 1: "yesemp", 2: "hard"}
elif model_name.lower() == "emptyJudge2".lower():
self.num_classes = 2
self.cls_dict = {0: "noemp", 1: "yesemp"}
self.model = VisionTransformer(config, self.args.img_size, zero_head=True, num_classes=self.num_classes, smoothing_value=self.args.smoothing_value)
if self.args.pretrained_model is not None:
self.model = torch.load(self.args.pretrained_model,map_location='cpu')
if not torch.cuda.is_available():
pretrained_model = torch.load(self.args.pretrained_model, map_location=torch.device('cpu'))['model']
self.model.load_state_dict(pretrained_model)
else:
pretrained_model = torch.load(self.args.pretrained_model)['model']
self.model.load_state_dict(pretrained_model)
self.model.to(self.args.device)
self.model.eval()
@ -61,7 +73,9 @@ class Predictor(object):
"Image file failed to read: {}".format(img_path))
else:
x = self.test_transform(img)
part_logits = self.model(x.unsqueeze(0).to(args.device))
if torch.cuda.is_available():
x = x.cuda()
part_logits = self.model(x.unsqueeze(0))
probs = torch.nn.Softmax(dim=-1)(part_logits)
topN = torch.argsort(probs, dim=-1, descending=True).tolist()
clas_ids = topN[0][0]
@ -75,12 +89,8 @@ if __name__ == "__main__":
y_true = []
y_pred = []
test_dir = "./emptyJudge5/images/"
test_dir = "/data/pfc/fineGrained/test_5cls"
dir_dict = {"noemp":"0", "yesemp":"1", "hard": "2", "fly": "3", "stack": "4"}
# test_dir = "../emptyJudge2/images"
# dir_dict = {"noempty":"0", "empty":"1"}
total = 0
num = 0
t0 = time.time()
@ -96,19 +106,6 @@ if __name__ == "__main__":
cur_pred, pred_score = predictor.normal_predict(cur_img_file)
label = 0 if 2 == int(label) or 3 == int(label) or 4 == int(label) else int(label)
riseresult = lightrise.riseempty(Image.open(cur_img_file))
if(label==1):
if(int(riseresult["rst_cls"])==1):
label=1
else:
label=0
# else:
# if(riseresult["rst_cls"]==0):
# label=0
# else:
# label=1
cur_pred = 0 if 2 == int(cur_pred) or 3 == int(cur_pred) or 4 == int(cur_pred) else int(cur_pred)
y_true.append(int(label))
y_pred.append(int(cur_pred))
@ -128,18 +125,6 @@ if __name__ == "__main__":
print(rst_C)
print(rst_f1)
'''
所有数据集
The cast of time is :160.738966 seconds
The classification accuracy is 0.986836
[[4923 58]
[ 34 1974]]
0.9839851634589902
'''
'''
test_imgs: yesemp=145, noemp=453 大图