JustinLin610
commited on
Commit
•
2f13d6c
1
Parent(s):
63b2141
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
---
|
4 |
+
|
5 |
+
# OFA-large-caption
|
6 |
+
This is the **large** version of OFA model finetuned for **image captioning**. OFA is a unified multimodal pretrained model that unifies modalities (i.e., cross-modality, vision, language) and tasks (e.g., image generation, visual grounding, image captioning, image classification, text generation, etc.) to a simple sequence-to-sequence learning framework.
|
7 |
+
|
8 |
+
The directory includes 4 files, namely `config.json` which consists of model configuration, `vocab.json` and `merge.txt` for our OFA tokenizer, and lastly `pytorch_model.bin` which consists of model weights. There is no need to worry about the mismatch between Fairseq and transformers, since we have addressed the issue yet.
|
9 |
+
|
10 |
+
To use it in transformers, please refer to https://github.com/OFA-Sys/OFA/tree/feature/add_transformers. Install the transformers and download the models as shown below.
|
11 |
+
```
|
12 |
+
git clone --single-branch --branch feature/add_transformers https://github.com/OFA-Sys/OFA.git
|
13 |
+
pip install OFA/transformers/
|
14 |
+
git clone https://huggingface.co/OFA-Sys/OFA-large-caption
|
15 |
+
```
|
16 |
+
After, refer the path to OFA-large to `ckpt_dir`, and prepare an image for the testing example below. Also, ensure that you have pillow and torchvision in your environment.
|
17 |
+
|
18 |
+
```
|
19 |
+
>>> from PIL import Image
|
20 |
+
>>> from torchvision import transforms
|
21 |
+
>>> from transformers import OFATokenizer, OFAModel
|
22 |
+
>>> from generate import sequence_generator
|
23 |
+
|
24 |
+
>>> mean, std = [0.5, 0.5, 0.5], [0.5, 0.5, 0.5]
|
25 |
+
>>> resolution = 480
|
26 |
+
>>> patch_resize_transform = transforms.Compose([
|
27 |
+
lambda image: image.convert("RGB"),
|
28 |
+
transforms.Resize((resolution, resolution), interpolation=Image.BICUBIC),
|
29 |
+
transforms.ToTensor(),
|
30 |
+
transforms.Normalize(mean=mean, std=std)
|
31 |
+
])
|
32 |
+
|
33 |
+
|
34 |
+
>>> tokenizer = OFATokenizer.from_pretrained(ckpt_dir)
|
35 |
+
|
36 |
+
>>> txt = " what does the image describe?"
|
37 |
+
>>> inputs = tokenizer([txt], return_tensors="pt").input_ids
|
38 |
+
>>> img = Image.open(path_to_image)
|
39 |
+
>>> patch_img = patch_resize_transform(img).unsqueeze(0)
|
40 |
+
|
41 |
+
|
42 |
+
>>> # using the generator of fairseq version
|
43 |
+
>>> model = OFAModel.from_pretrained(ckpt_dir, use_cache=True)
|
44 |
+
>>> generator = sequence_generator.SequenceGenerator(
|
45 |
+
tokenizer=tokenizer,
|
46 |
+
beam_size=5,
|
47 |
+
max_len_b=16,
|
48 |
+
min_len=0,
|
49 |
+
no_repeat_ngram_size=3,
|
50 |
+
)
|
51 |
+
>>> data = {}
|
52 |
+
>>> data["net_input"] = {"input_ids": inputs, 'patch_images': patch_img, 'patch_masks':torch.tensor([True])}
|
53 |
+
>>> gen_output = generator.generate([model], data)
|
54 |
+
>>> gen = [gen_output[i][0]["tokens"] for i in range(len(gen_output))]
|
55 |
+
|
56 |
+
>>> # using the generator of huggingface version
|
57 |
+
>>> model = OFAModel.from_pretrained(ckpt_dir, use_cache=False)
|
58 |
+
>>> gen = model.generate(inputs, patch_images=patch_img, num_beams=5, no_repeat_ngram_size=3)
|
59 |
+
|
60 |
+
>>> print(tokenizer.batch_decode(gen, skip_special_tokens=True))
|
61 |
+
```
|