Add evaluation script
Browse files
README.md
CHANGED
@@ -27,6 +27,7 @@ git clone https://github.com/hmichaeli/alias_free_convnets.git
|
|
27 |
```python
|
28 |
from huggingface_hub import hf_hub_download
|
29 |
import torch
|
|
|
30 |
from alias_free_convnets.models.convnext_afc import convnext_afc_tiny
|
31 |
|
32 |
# baseline
|
@@ -51,6 +52,56 @@ afc_model = convnext_afc_tiny(
|
|
51 |
)
|
52 |
afc_model.load_state_dict(ckpt, strict=False)
|
53 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
```
|
55 |
|
56 |
|
|
|
27 |
```python
|
28 |
from huggingface_hub import hf_hub_download
|
29 |
import torch
|
30 |
+
from torchvision import datasets, transforms
|
31 |
from alias_free_convnets.models.convnext_afc import convnext_afc_tiny
|
32 |
|
33 |
# baseline
|
|
|
52 |
)
|
53 |
afc_model.load_state_dict(ckpt, strict=False)
|
54 |
|
55 |
+
# evaluate model
|
56 |
+
interpolation = transforms.InterpolationMode.BICUBIC
|
57 |
+
IMAGENET_DEFAULT_MEAN = (0.485, 0.456, 0.406)
|
58 |
+
IMAGENET_DEFAULT_STD = (0.229, 0.224, 0.225)
|
59 |
+
transform = transforms.Compose([
|
60 |
+
transforms.Resize(256, interpolation=interpolation),
|
61 |
+
transforms.CenterCrop(224),
|
62 |
+
transforms.ToTensor(),
|
63 |
+
transforms.Normalize(IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD),
|
64 |
+
])
|
65 |
+
data_path = "/path/to/imagenet/val"
|
66 |
+
dataset_val = datasets.ImageFolder(data_path, transform=transform)
|
67 |
+
nb_classes = 1000
|
68 |
+
sampler_val = torch.utils.data.SequentialSampler(dataset_val)
|
69 |
+
data_loader_val = torch.utils.data.DataLoader(
|
70 |
+
dataset_val, sampler=sampler_val,
|
71 |
+
batch_size=8,
|
72 |
+
num_workers=8,
|
73 |
+
drop_last=False
|
74 |
+
)
|
75 |
+
|
76 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
77 |
+
|
78 |
+
@torch.no_grad()
|
79 |
+
def evaluate(data_loader, model, device):
|
80 |
+
model.eval()
|
81 |
+
correct = 0
|
82 |
+
total = 0
|
83 |
+
for batch_idx, (inputs, targets) in enumerate(data_loader):
|
84 |
+
inputs, targets = inputs.to(device), targets.to(device)
|
85 |
+
outputs = model(inputs)
|
86 |
+
_, predicted = outputs.max(1)
|
87 |
+
total += targets.size(0)
|
88 |
+
correct += predicted.eq(targets).sum().item()
|
89 |
+
|
90 |
+
acc = 100. * correct / total
|
91 |
+
print("Acc@1 {:.3f}".format(acc))
|
92 |
+
|
93 |
+
|
94 |
+
|
95 |
+
print("evaluate baseline")
|
96 |
+
base_model.to(device)
|
97 |
+
test_stats = evaluate(data_loader_val, base_model, device)
|
98 |
+
|
99 |
+
print("evaluate AFC")
|
100 |
+
afc_model.to(device)
|
101 |
+
test_stats = evaluate(data_loader_val, afc_model, device)
|
102 |
+
|
103 |
+
|
104 |
+
|
105 |
```
|
106 |
|
107 |
|