Delete Transformer-Explainability/dataset
Browse files
Transformer-Explainability/dataset/expl_hdf5.py
DELETED
@@ -1,52 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
|
3 |
-
import h5py
|
4 |
-
import torch
|
5 |
-
from torch.utils.data import Dataset
|
6 |
-
|
7 |
-
|
8 |
-
class ImagenetResults(Dataset):
|
9 |
-
def __init__(self, path):
|
10 |
-
super(ImagenetResults, self).__init__()
|
11 |
-
|
12 |
-
self.path = os.path.join(path, "results.hdf5")
|
13 |
-
self.data = None
|
14 |
-
|
15 |
-
print("Reading dataset length...")
|
16 |
-
with h5py.File(self.path, "r") as f:
|
17 |
-
# tmp = h5py.File(self.path , 'r')
|
18 |
-
self.data_length = len(f["/image"])
|
19 |
-
|
20 |
-
def __len__(self):
|
21 |
-
return self.data_length
|
22 |
-
|
23 |
-
def __getitem__(self, item):
|
24 |
-
if self.data is None:
|
25 |
-
self.data = h5py.File(self.path, "r")
|
26 |
-
|
27 |
-
image = torch.tensor(self.data["image"][item])
|
28 |
-
vis = torch.tensor(self.data["vis"][item])
|
29 |
-
target = torch.tensor(self.data["target"][item]).long()
|
30 |
-
|
31 |
-
return image, vis, target
|
32 |
-
|
33 |
-
|
34 |
-
if __name__ == "__main__":
|
35 |
-
import imageio
|
36 |
-
import numpy as np
|
37 |
-
from utils import render
|
38 |
-
|
39 |
-
ds = ImagenetResults("../visualizations/fullgrad")
|
40 |
-
sample_loader = torch.utils.data.DataLoader(ds, batch_size=5, shuffle=False)
|
41 |
-
|
42 |
-
iterator = iter(sample_loader)
|
43 |
-
image, vis, target = next(iterator)
|
44 |
-
|
45 |
-
maps = (
|
46 |
-
render.hm_to_rgb(vis[0].data.cpu().numpy(), scaling=3, sigma=1, cmap="seismic")
|
47 |
-
* 255
|
48 |
-
).astype(np.uint8)
|
49 |
-
|
50 |
-
# imageio.imsave('../delete_hm.jpg', maps)
|
51 |
-
|
52 |
-
print(len(ds))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|