Update README.md
Browse files
README.md
CHANGED
|
@@ -252,13 +252,21 @@ The following hyperparameters were used during training:
|
|
| 252 |
|
| 253 |
### Example of usage
|
| 254 |
|
|
|
|
|
|
|
| 255 |
```python
|
| 256 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 257 |
from transformers import TrainingArguments
|
|
|
|
| 258 |
from transformers import CLIPProcessor, AutoModelForImageClassification
|
|
|
|
| 259 |
|
| 260 |
processor = CLIPProcessor.from_pretrained("Andron00e/CLIPForImageClassification-v1")
|
| 261 |
-
model =
|
| 262 |
|
| 263 |
dataset = load_dataset("Andron00e/CIFAR10-custom")
|
| 264 |
dataset = dataset["train"].train_test_split(test_size=0.2)
|
|
@@ -271,6 +279,8 @@ dataset = DatasetDict({
|
|
| 271 |
"test": val_test["test"],
|
| 272 |
})
|
| 273 |
|
|
|
|
|
|
|
| 274 |
def transform(example_batch):
|
| 275 |
inputs = processor(text=[classes[x] for x in example_batch['labels']], images=[x for x in example_batch['image']], padding=True, return_tensors='pt')
|
| 276 |
inputs['labels'] = example_batch['labels']
|
|
@@ -284,6 +294,11 @@ def collate_fn(batch):
|
|
| 284 |
'labels': torch.tensor([x['labels'] for x in batch])
|
| 285 |
}
|
| 286 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 287 |
training_args = TrainingArguments(
|
| 288 |
output_dir="./outputs",
|
| 289 |
per_device_train_batch_size=16,
|
|
@@ -322,4 +337,7 @@ trainer.save_state()
|
|
| 322 |
metrics = trainer.evaluate(processed_dataset['test'])
|
| 323 |
trainer.log_metrics("eval", metrics)
|
| 324 |
trainer.save_metrics("eval", metrics)
|
|
|
|
|
|
|
|
|
|
| 325 |
```
|
|
|
|
| 252 |
|
| 253 |
### Example of usage
|
| 254 |
|
| 255 |
+
Simple demo for Google Colab
|
| 256 |
+
|
| 257 |
```python
|
| 258 |
+
!pip install datasets transformers[torch] accelerate -U
|
| 259 |
+
!git clone https://github.com/Andron00e/CLIPForImageClassification
|
| 260 |
+
%cd CLIPForImageClassification/clip_for_classification
|
| 261 |
+
|
| 262 |
+
import torch
|
| 263 |
from transformers import TrainingArguments
|
| 264 |
+
from datasets import load_dataset, load_metric
|
| 265 |
from transformers import CLIPProcessor, AutoModelForImageClassification
|
| 266 |
+
from modeling_clipforimageclassification import CLIPForImageClassification
|
| 267 |
|
| 268 |
processor = CLIPProcessor.from_pretrained("Andron00e/CLIPForImageClassification-v1")
|
| 269 |
+
model = CLIPForImageClassification.from_pretrained("Andron00e/CLIPForImageClassification-v1", 10)
|
| 270 |
|
| 271 |
dataset = load_dataset("Andron00e/CIFAR10-custom")
|
| 272 |
dataset = dataset["train"].train_test_split(test_size=0.2)
|
|
|
|
| 279 |
"test": val_test["test"],
|
| 280 |
})
|
| 281 |
|
| 282 |
+
classes = {0: "airplane", 1: "automobile", 2: "bird", 3: "cat", 4: "deer", 5: "dog", 6: "frog", 7: "horse", 8: "ship", 9: "truck"}
|
| 283 |
+
|
| 284 |
def transform(example_batch):
|
| 285 |
inputs = processor(text=[classes[x] for x in example_batch['labels']], images=[x for x in example_batch['image']], padding=True, return_tensors='pt')
|
| 286 |
inputs['labels'] = example_batch['labels']
|
|
|
|
| 294 |
'labels': torch.tensor([x['labels'] for x in batch])
|
| 295 |
}
|
| 296 |
|
| 297 |
+
metric = load_metric("accuracy")
|
| 298 |
+
|
| 299 |
+
def compute_metrics(p):
|
| 300 |
+
return metric.compute(predictions=np.argmax(p.predictions, axis=1), references=p.label_ids)
|
| 301 |
+
|
| 302 |
training_args = TrainingArguments(
|
| 303 |
output_dir="./outputs",
|
| 304 |
per_device_train_batch_size=16,
|
|
|
|
| 337 |
metrics = trainer.evaluate(processed_dataset['test'])
|
| 338 |
trainer.log_metrics("eval", metrics)
|
| 339 |
trainer.save_metrics("eval", metrics)
|
| 340 |
+
|
| 341 |
+
%cd ..
|
| 342 |
+
%cd ..
|
| 343 |
```
|