-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathfew_shot_finetuning.py
166 lines (132 loc) · 6.64 KB
/
few_shot_finetuning.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
from pathlib import Path
import random
import clip
import numpy as np
import torch
from torch.utils.tensorboard import SummaryWriter
import torchmetrics
import torchvision
from tqdm import tqdm
from balanced_batch_sampler import BalancedBatchSampler
def finetune():
torch.manual_seed(0)
random.seed(0)
np.random.seed(0)
SAVE_INTERVAL = 10
BATCH_SIZE = 8
NUM_EPOCHS = 100
def convert_models_to_fp32(model):
for p in model.parameters():
p.data = p.data.float()
if p.requires_grad:
p.grad.data = p.grad.data.float()
device = "cuda:0" if torch.cuda.is_available() else "cpu"
model, preprocess = clip.load("ViT-B/32", device=device, jit=False) #Must set jit=False for training
if device == "cpu":
model.float()
else:
clip.model.convert_weights(model) # Actually this line is unnecessary since clip by default already on float16
writer = SummaryWriter()
weights_path = Path("model_checkpoints")
weights_path.mkdir(exist_ok=True)
train_dataset = torchvision.datasets.ImageFolder("data/coco_crops_few_shot/train", transform=preprocess)
train_labels = torch.tensor(train_dataset.targets)
train_sampler = BalancedBatchSampler(train_labels, BATCH_SIZE, 1)
# use drop_last = True to ensure each batch contains 8 target classes to choose from.
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_sampler=train_sampler,
drop_last=True)
test_dataset = torchvision.datasets.ImageFolder("data/coco_crops_few_shot/test", transform=preprocess)
test_labels = torch.tensor(test_dataset.targets)
test_sampler = BalancedBatchSampler(test_labels, BATCH_SIZE, 1)
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_sampler=test_sampler,
drop_last=True)
loss_img = torch.nn.CrossEntropyLoss()
loss_txt = torch.nn.CrossEntropyLoss()
# for p in model.transformer.parameters():
# p.requires_grad = False
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.Adam(
params, lr=1e-7, weight_decay=0.0001)
num_batches_train = len(train_dataloader.dataset)/BATCH_SIZE
num_batches_val = len(test_dataloader.dataset)/BATCH_SIZE
for epoch in range(NUM_EPOCHS):
print(f"Epoch: {epoch}")
epoch_train_loss = 0
model.train()
for batch in tqdm(train_dataloader,total=num_batches_train):
optimizer.zero_grad()
images, class_ids = batch
images = torch.stack([img for img in images], dim=0).to(
device
)
# TODO: to use mean of multiple prompts need to pre-compute them.
texts = [f"a photo of a {train_dataloader.dataset.classes[label_id]}" for label_id in class_ids]
texts = clip.tokenize(texts).to(device)
logits_per_image, logits_per_text = model(images, texts)
ground_truth = torch.arange(logits_per_image.shape[0], dtype=torch.long, device=device)
total_train_loss = (loss_img(logits_per_image, ground_truth) + loss_txt(logits_per_text, ground_truth)) / 2
total_train_loss.backward()
epoch_train_loss += total_train_loss
torch.nn.utils.clip_grad_norm_(params, 1.0)
if device == "cpu":
optimizer.step()
else:
convert_models_to_fp32(model)
optimizer.step()
clip.model.convert_weights(model)
epoch_train_loss /= num_batches_train
writer.add_scalar("Loss/train", epoch_train_loss, epoch)
if epoch % SAVE_INTERVAL == 0:
torch.save(
{
'epoch': epoch,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'loss': epoch_train_loss,
}, weights_path / f"model_{epoch}.pt") #just change to your preferred folder/filename
print(f"Saved weights under model_checkpoint/model_{epoch}.pt.")
# Compute test accuracy
model.eval()
values_list, indices_list = [], []
top5_results = []
top1_results = []
acc_top1_list = []
acc_top5_list = []
num_batches_test = len(test_dataloader.dataset)/BATCH_SIZE
epoch_test_loss = 0
for i, batch in enumerate(tqdm(test_dataloader, total=num_batches_test)):
images, class_ids = batch
class_ids = class_ids.to(device)
images = images.to(device)
texts = torch.cat([clip.tokenize(f"a photo of a {c}") for c in test_dataloader.dataset.classes]).to(device)
with torch.no_grad():
# TODO: remove duplicate computation of image and text features
image_features = model.encode_image(images)
text_features = model.encode_text(texts)
logits_per_image, logits_per_text = model(images, texts)
ground_truth = torch.arange(logits_per_image.shape[0], dtype=torch.long, device=device)
total_loss = (loss_img(logits_per_image, ground_truth) + loss_txt(logits_per_text, ground_truth)) / 2
epoch_test_loss += total_loss
image_features /= image_features.norm(dim=-1, keepdim=True)
text_features /= text_features.norm(dim=-1, keepdim=True)
similarity = (100.0 * image_features @ text_features.T).softmax(dim=-1)
acc_top1 = torchmetrics.functional.accuracy(similarity, class_ids)
acc_top5 = torchmetrics.functional.accuracy(similarity, class_ids, top_k=5)
acc_top1_list.append(acc_top1)
acc_top5_list.append(acc_top5)
writer.add_scalar("Loss/test", epoch_test_loss / num_batches_test, epoch)
print(f"Epoch {epoch} train loss: {epoch_train_loss / num_batches_train}")
print(f"Epoch {epoch} test loss: {epoch_test_loss / num_batches_test}")
# compute mean top5 accuracy and top1 accuracy
mean_top5_accuracy = torch.stack(acc_top5_list).mean().cpu().numpy()
print(f"Mean Top 5 Accuracy: {mean_top5_accuracy*100}%.")
writer.add_scalar("Test Accuracy/Top5", mean_top5_accuracy , epoch)
mean_top1_accuracy = torch.stack(acc_top1_list).mean().cpu().numpy()
print(f"Mean Top 1 Accuracy: {mean_top1_accuracy*100}%.")
writer.add_scalar("Test Accuracy/Top1", mean_top1_accuracy, epoch)
writer.flush()
writer.close()
if __name__ == '__main__':
finetune()