Spaces:
Running
Running
Upload configs/retrieval_flickr.yaml
Browse files
configs/retrieval_flickr.yaml
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
image_root: '/export/share/datasets/vision/flickr30k/'
|
2 |
+
ann_root: 'annotation'
|
3 |
+
dataset: 'flickr'
|
4 |
+
|
5 |
+
# set pretrained as a file path or an url
|
6 |
+
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_flickr.pth'
|
7 |
+
|
8 |
+
# size of vit model; base or large
|
9 |
+
|
10 |
+
vit: 'base'
|
11 |
+
batch_size_train: 32
|
12 |
+
batch_size_test: 64
|
13 |
+
vit_grad_ckpt: True
|
14 |
+
vit_ckpt_layer: 4
|
15 |
+
init_lr: 1e-5
|
16 |
+
|
17 |
+
# vit: 'large'
|
18 |
+
# batch_size_train: 16
|
19 |
+
# batch_size_test: 32
|
20 |
+
# vit_grad_ckpt: True
|
21 |
+
# vit_ckpt_layer: 10
|
22 |
+
# init_lr: 5e-6
|
23 |
+
|
24 |
+
image_size: 384
|
25 |
+
queue_size: 57600
|
26 |
+
alpha: 0.4
|
27 |
+
k_test: 128
|
28 |
+
negative_all_rank: False
|
29 |
+
|
30 |
+
# optimizer
|
31 |
+
weight_decay: 0.05
|
32 |
+
min_lr: 0
|
33 |
+
max_epoch: 6
|
34 |
+
|