Commit
•
039f477
1
Parent(s):
af83b6b
Upload folder using huggingface_hub
Browse files- .gitattributes +0 -1
- README.md +128 -0
- config.json +40 -0
- image_encoder.mlpackage/Data/com.apple.CoreML/model.mlmodel +3 -0
- image_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin +3 -0
- image_encoder.mlpackage/Manifest.json +18 -0
- image_encoder.onnx +3 -0
- image_encoder.pt +3 -0
- image_encoder.safetensors +3 -0
- text_encoder.mlpackage/Data/com.apple.CoreML/model.mlmodel +3 -0
- text_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin +3 -0
- text_encoder.mlpackage/Manifest.json +18 -0
- text_encoder.onnx +3 -0
- text_encoder.pt +3 -0
- text_encoder.safetensors +3 -0
- tokenizer.json +0 -0
.gitattributes
CHANGED
@@ -25,7 +25,6 @@
|
|
25 |
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
28 |
-
*.tar filter=lfs diff=lfs merge=lfs -text
|
29 |
*.tflite filter=lfs diff=lfs merge=lfs -text
|
30 |
*.tgz filter=lfs diff=lfs merge=lfs -text
|
31 |
*.wasm filter=lfs diff=lfs merge=lfs -text
|
|
|
25 |
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
|
|
28 |
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
*.wasm filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
pipeline_tag: feature-extraction
|
4 |
+
tags:
|
5 |
+
- clip
|
6 |
+
- vision
|
7 |
+
datasets:
|
8 |
+
- sbu_captions
|
9 |
+
- visual_genome
|
10 |
+
- ChristophSchuhmann/MS_COCO_2017_URL_TEXT
|
11 |
+
---
|
12 |
+
<h1 align="center">UForm</h1>
|
13 |
+
<h3 align="center">
|
14 |
+
Multi-Modal Inference Library<br/>
|
15 |
+
For Semantic Search Applications<br/>
|
16 |
+
</h3>
|
17 |
+
|
18 |
+
---
|
19 |
+
|
20 |
+
UForm is a Multi-Modal Modal Inference package, designed to encode Multi-Lingual Texts, Images, and, soon, Audio, Video, and Documents, into a shared vector space!
|
21 |
+
|
22 |
+
This is model card of the __English only model__ with:
|
23 |
+
|
24 |
+
* 4 layers BERT (2 layers for unimodal encoding and rest layers for multimodal encoding)
|
25 |
+
* ViT-B/16 (image resolution is 224x224)
|
26 |
+
|
27 |
+
|
28 |
+
If you need Multilingual model, check [this](https://huggingface.co/unum-cloud/uform-vl-multilingual).
|
29 |
+
|
30 |
+
## Evaluation
|
31 |
+
|
32 |
+
The following metrics were obtained with multimodal re-ranking:
|
33 |
+
|
34 |
+
| Dataset | Recall@1 | Recall@5 | Recall@10 |
|
35 |
+
| :-------- | ------: | --------: | --------: |
|
36 |
+
| Zero-Shot Flickr | 0.727 | 0.915 | 0.949 |
|
37 |
+
| MS-COCO (train split was in training data) | 0.510 | 0.761 | 0.838 |
|
38 |
+
|
39 |
+
## Installation
|
40 |
+
|
41 |
+
```bash
|
42 |
+
pip install uform[torch]
|
43 |
+
```
|
44 |
+
|
45 |
+
## Usage
|
46 |
+
|
47 |
+
To load the model:
|
48 |
+
|
49 |
+
```python
|
50 |
+
import uform
|
51 |
+
|
52 |
+
model, processor = uform.get_model('unum-cloud/uform-vl-english')
|
53 |
+
```
|
54 |
+
|
55 |
+
To encode data:
|
56 |
+
|
57 |
+
```python
|
58 |
+
from PIL import Image
|
59 |
+
|
60 |
+
text = 'a small red panda in a zoo'
|
61 |
+
image = Image.open('red_panda.jpg')
|
62 |
+
|
63 |
+
image_data = processor.preprocess_image(image)
|
64 |
+
text_data = processor.preprocess_text(text)
|
65 |
+
|
66 |
+
image_features, image_embedding = model.encode_image(image_data, return_features=True)
|
67 |
+
text_features, text_embedding = model.encode_text(text_data, return_features=True)
|
68 |
+
```
|
69 |
+
|
70 |
+
To get features:
|
71 |
+
|
72 |
+
```python
|
73 |
+
image_features, image_embedding = model.encode_image(image_data, return_features=True)
|
74 |
+
text_features, text_embedding = model.encode_text(text_data, return_features=True)
|
75 |
+
```
|
76 |
+
|
77 |
+
These features can later be used to produce joint multimodal encodings faster, as the first layers of the transformer can be skipped:
|
78 |
+
|
79 |
+
```python
|
80 |
+
joint_embedding = model.encode_multimodal(
|
81 |
+
image_features=image_features,
|
82 |
+
text_features=text_features,
|
83 |
+
attention_mask=text_data['attention_mask']
|
84 |
+
)
|
85 |
+
```
|
86 |
+
|
87 |
+
There are two options to calculate semantic compatibility between an image and a text: [Cosine Similarity](#cosine-similarity) and [Matching Score](#matching-score).
|
88 |
+
|
89 |
+
### Cosine Similarity
|
90 |
+
|
91 |
+
```python
|
92 |
+
import torch.nn.functional as F
|
93 |
+
|
94 |
+
similarity = F.cosine_similarity(image_embedding, text_embedding)
|
95 |
+
```
|
96 |
+
|
97 |
+
The `similarity` will belong to the `[-1, 1]` range, `1` meaning the absolute match.
|
98 |
+
|
99 |
+
__Pros__:
|
100 |
+
|
101 |
+
- Computationally cheap.
|
102 |
+
- Only unimodal embeddings are required, unimodal encoding is faster than joint encoding.
|
103 |
+
- Suitable for retrieval in large collections.
|
104 |
+
|
105 |
+
__Cons__:
|
106 |
+
|
107 |
+
- Takes into account only coarse-grained features.
|
108 |
+
|
109 |
+
|
110 |
+
### Matching Score
|
111 |
+
|
112 |
+
Unlike cosine similarity, unimodal embedding are not enough.
|
113 |
+
Joint embedding will be needed and the resulting `score` will belong to the `[0, 1]` range, `1` meaning the absolute match.
|
114 |
+
|
115 |
+
```python
|
116 |
+
score = model.get_matching_scores(joint_embedding)
|
117 |
+
```
|
118 |
+
|
119 |
+
__Pros__:
|
120 |
+
|
121 |
+
- Joint embedding captures fine-grained features.
|
122 |
+
- Suitable for re-ranking – sorting retrieval result.
|
123 |
+
|
124 |
+
__Cons__:
|
125 |
+
|
126 |
+
- Resource-intensive.
|
127 |
+
- Not suitable for retrieval in large collections.
|
128 |
+
|
config.json
ADDED
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"text_encoder": {
|
3 |
+
"tokenizer_class": "bert",
|
4 |
+
"model_type": "bert",
|
5 |
+
"dim": 768,
|
6 |
+
"context_dim": 768,
|
7 |
+
"vocab_size": 30522,
|
8 |
+
"padding_idx": 0,
|
9 |
+
"num_layers": 4,
|
10 |
+
"num_heads": 12,
|
11 |
+
"embedding_dim": 256,
|
12 |
+
"multimodal_layers_ids": [
|
13 |
+
2,
|
14 |
+
3
|
15 |
+
],
|
16 |
+
"head_one_neuron": false,
|
17 |
+
"pooling": "cls",
|
18 |
+
"max_position_embeddings": 77,
|
19 |
+
"dropout_prob": 0.1
|
20 |
+
},
|
21 |
+
"image_encoder": {
|
22 |
+
"normalization_means": [
|
23 |
+
0.48145466,
|
24 |
+
0.4578275,
|
25 |
+
0.40821073
|
26 |
+
],
|
27 |
+
"normalization_deviations": [
|
28 |
+
0.26862954,
|
29 |
+
0.26130258,
|
30 |
+
0.27577711
|
31 |
+
],
|
32 |
+
"dim": 768,
|
33 |
+
"patch_size": 16,
|
34 |
+
"image_size": 224,
|
35 |
+
"num_layers": 12,
|
36 |
+
"num_heads": 12,
|
37 |
+
"embedding_dim": 256,
|
38 |
+
"pooling": "cls"
|
39 |
+
}
|
40 |
+
}
|
image_encoder.mlpackage/Data/com.apple.CoreML/model.mlmodel
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:101d2a485d7f1fb9b13bb2e9b21017d41e0484c09d7252fa48cd45663bf6a7a0
|
3 |
+
size 111239
|
image_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d795737604c1677ff6b2e5fa8bceface13912a9baecea99b6ef3e24efa050464
|
3 |
+
size 344067136
|
image_encoder.mlpackage/Manifest.json
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"fileFormatVersion": "1.0.0",
|
3 |
+
"itemInfoEntries": {
|
4 |
+
"ADEDA51D-BFFF-46BC-AF60-BF0FB2F3BB19": {
|
5 |
+
"author": "com.apple.CoreML",
|
6 |
+
"description": "CoreML Model Specification",
|
7 |
+
"name": "model.mlmodel",
|
8 |
+
"path": "com.apple.CoreML/model.mlmodel"
|
9 |
+
},
|
10 |
+
"F1FDEC64-5002-491E-8C51-99A28404DD54": {
|
11 |
+
"author": "com.apple.CoreML",
|
12 |
+
"description": "CoreML Model Weights",
|
13 |
+
"name": "weights",
|
14 |
+
"path": "com.apple.CoreML/weights"
|
15 |
+
}
|
16 |
+
},
|
17 |
+
"rootModelIdentifier": "ADEDA51D-BFFF-46BC-AF60-BF0FB2F3BB19"
|
18 |
+
}
|
image_encoder.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:54ea2e32abb19d5a8415143f3d2d87ce3e96ccc0a031653af198db6fd13ecef8
|
3 |
+
size 87267681
|
image_encoder.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ea4548002193d407577533b6a98596a9903f9576c4c8de574b06050aaa5c92d7
|
3 |
+
size 172100742
|
image_encoder.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:800f75efc508c0eb3445ec270152194e692860656eef32b8ebab888ff5946351
|
3 |
+
size 172047072
|
text_encoder.mlpackage/Data/com.apple.CoreML/model.mlmodel
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f1428abd50aa8fe334c81f5bc88c8fb62b9903ba05ad94e61d0f6d73f3cf5c42
|
3 |
+
size 21788
|
text_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8dd08b0a4867b25785d44b8dce1c37174894c788a732dd18c2ab3f8b47f048d3
|
3 |
+
size 151499200
|
text_encoder.mlpackage/Manifest.json
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"fileFormatVersion": "1.0.0",
|
3 |
+
"itemInfoEntries": {
|
4 |
+
"546A7C3A-6CCD-42C4-A20B-507DBA62CEA6": {
|
5 |
+
"author": "com.apple.CoreML",
|
6 |
+
"description": "CoreML Model Specification",
|
7 |
+
"name": "model.mlmodel",
|
8 |
+
"path": "com.apple.CoreML/model.mlmodel"
|
9 |
+
},
|
10 |
+
"D48DD071-5007-4390-BF8E-9D05D8483749": {
|
11 |
+
"author": "com.apple.CoreML",
|
12 |
+
"description": "CoreML Model Weights",
|
13 |
+
"name": "weights",
|
14 |
+
"path": "com.apple.CoreML/weights"
|
15 |
+
}
|
16 |
+
},
|
17 |
+
"rootModelIdentifier": "546A7C3A-6CCD-42C4-A20B-507DBA62CEA6"
|
18 |
+
}
|
text_encoder.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ac4e7765b4951b4ebe9001f35d7deb1dadf77cde697c363ba90d2c44f71c052b
|
3 |
+
size 38004782
|
text_encoder.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:50bc23ed1f2642decba5e2bc7eca5bc9d83cdc7e13f44fd00298395d90849cb3
|
3 |
+
size 113589216
|
text_encoder.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:7001e423ec999a2bfbd33683f48f8163942d3e9a7259944db170129e0857ffdf
|
3 |
+
size 113566956
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|