Delete script.py
Browse files
script.py
DELETED
@@ -1,90 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
import pandas as pd
|
3 |
-
import torch
|
4 |
-
from PIL import Image
|
5 |
-
from torchvision import transforms
|
6 |
-
|
7 |
-
from model import efficientnetv2_s as create_model
|
8 |
-
|
9 |
-
|
10 |
-
def predict(test_metadata, root_path='/tmp/data/private_testset', output_csv_path='./submission.csv'):
|
11 |
-
|
12 |
-
img_size = {"s": [384, 384], # train_size, val_size
|
13 |
-
"m": [384, 480],
|
14 |
-
"l": [384, 480]}
|
15 |
-
num_model = "s"
|
16 |
-
|
17 |
-
data_transform = transforms.Compose(
|
18 |
-
[transforms.Resize(img_size[num_model][1]),
|
19 |
-
transforms.CenterCrop(img_size[num_model][1]),
|
20 |
-
transforms.ToTensor(),
|
21 |
-
transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])])
|
22 |
-
|
23 |
-
id_list = test_metadata['observation_id'].tolist()
|
24 |
-
img_name_list = test_metadata['filename'].tolist()
|
25 |
-
print(os.path.abspath(os.path.dirname(__file__)))
|
26 |
-
|
27 |
-
id2classId = dict()
|
28 |
-
id2prob = dict()
|
29 |
-
prob_list = list()
|
30 |
-
classId_list = list()
|
31 |
-
|
32 |
-
for img_name in img_name_list:
|
33 |
-
img_path = os.path.join(root_path, img_name)
|
34 |
-
assert os.path.exists(img_path), "file: '{}' dose not exist.".format(img_path)
|
35 |
-
img = Image.open(img_path).convert('RGB')
|
36 |
-
img = data_transform(img)
|
37 |
-
img = torch.unsqueeze(img, dim=0)
|
38 |
-
|
39 |
-
with torch.no_grad():
|
40 |
-
# predict class
|
41 |
-
output = model(img.to(device)).cpu()
|
42 |
-
predict = torch.softmax(output, dim=1)
|
43 |
-
probs, classesId = torch.max(predict, dim=1)
|
44 |
-
prob = probs.data.numpy().tolist()[0]
|
45 |
-
classesId = classesId.data.numpy().tolist()[0]
|
46 |
-
prob_list.append(prob)
|
47 |
-
classId_list.append(classesId)
|
48 |
-
|
49 |
-
for i, id in enumerate(id_list):
|
50 |
-
if id not in id2classId.keys():
|
51 |
-
id2classId[id] = classId_list[i]
|
52 |
-
id2prob[id] = prob_list[i]
|
53 |
-
else:
|
54 |
-
if prob_list[i] > id2prob[id]:
|
55 |
-
id2classId[id] = classId_list[i]
|
56 |
-
id2prob[id] = prob_list[i]
|
57 |
-
classes = list()
|
58 |
-
for id in id_list:
|
59 |
-
classes.append(str(id2classId[id]))
|
60 |
-
test_metadata["class_id"] = classes
|
61 |
-
|
62 |
-
user_pred_df = test_metadata.drop_duplicates("observation_id", keep="first")
|
63 |
-
user_pred_df[["observation_id", "class_id"]].to_csv(output_csv_path, index=None)
|
64 |
-
|
65 |
-
|
66 |
-
if __name__ == '__main__':
|
67 |
-
import zipfile
|
68 |
-
|
69 |
-
with zipfile.ZipFile("/tmp/data/private_testset.zip", 'r') as zip_ref:
|
70 |
-
zip_ref.extractall("/tmp/data")
|
71 |
-
root_path = '/tmp/data/private_testset'
|
72 |
-
|
73 |
-
# root_path = "../../data_set/flower_data/val/n1"
|
74 |
-
|
75 |
-
# json_file = open(json_path, "r")
|
76 |
-
# index2class = json.load(json_file)
|
77 |
-
|
78 |
-
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
79 |
-
# create model
|
80 |
-
model = create_model(num_classes=1784).to(device)
|
81 |
-
|
82 |
-
# load model weights
|
83 |
-
model_weight_path = "./efficientNetV2.pth"
|
84 |
-
model.load_state_dict(torch.load(model_weight_path, map_location=device))
|
85 |
-
model.eval()
|
86 |
-
|
87 |
-
metadata_file_path = "./SnakeCLEF2024_TestMetadata.csv"
|
88 |
-
# metadata_file_path = "./test1.csv"
|
89 |
-
test_metadata = pd.read_csv(metadata_file_path)
|
90 |
-
predict(test_metadata, root_path)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|