Spaces:
Running
on
Zero
Running
on
Zero
ahmed-masry
commited on
Commit
•
e8ad0b4
1
Parent(s):
9194fb8
Update processing_colflor.py
Browse files- processing_colflor.py +80 -80
processing_colflor.py
CHANGED
@@ -1,81 +1,81 @@
|
|
1 |
-
from typing import List, Optional, Union
|
2 |
-
|
3 |
-
import torch
|
4 |
-
from PIL import Image
|
5 |
-
from transformers import BatchFeature
|
6 |
-
|
7 |
-
from
|
8 |
-
|
9 |
-
from colpali_engine.utils.processing_utils import BaseVisualRetrieverProcessor
|
10 |
-
|
11 |
-
|
12 |
-
class ColFlorProcessor(BaseVisualRetrieverProcessor, Florence2Processor):
|
13 |
-
"""
|
14 |
-
Processor for ColPali.
|
15 |
-
"""
|
16 |
-
|
17 |
-
def __init__(self, *args, **kwargs):
|
18 |
-
super().__init__(*args, **kwargs)
|
19 |
-
self.mock_image = Image.new("RGB", (16, 16), color="black")
|
20 |
-
|
21 |
-
def process_images(
|
22 |
-
self,
|
23 |
-
images: List[Image.Image],
|
24 |
-
) -> BatchFeature:
|
25 |
-
"""
|
26 |
-
Process images for ColFlor2.
|
27 |
-
"""
|
28 |
-
texts_doc = ["<OCR>"] * len(images)
|
29 |
-
images = [image.convert("RGB") for image in images]
|
30 |
-
|
31 |
-
batch_doc = self(
|
32 |
-
text=texts_doc,
|
33 |
-
images=images,
|
34 |
-
return_tensors="pt",
|
35 |
-
padding="longest",
|
36 |
-
)
|
37 |
-
|
38 |
-
new_part = torch.ones((batch_doc['attention_mask'].size()[0], 577)).to(batch_doc['attention_mask'].device)
|
39 |
-
batch_doc['full_attention_mask'] = torch.cat([new_part, batch_doc['attention_mask']], dim=1)
|
40 |
-
|
41 |
-
return batch_doc
|
42 |
-
|
43 |
-
def process_queries(
|
44 |
-
self,
|
45 |
-
queries: List[str],
|
46 |
-
max_length: int = 50,
|
47 |
-
suffix: Optional[str] = None,
|
48 |
-
) -> BatchFeature:
|
49 |
-
"""
|
50 |
-
Process queries for ColFlor2.
|
51 |
-
"""
|
52 |
-
if suffix is None:
|
53 |
-
suffix = "<pad>" * 10
|
54 |
-
texts_query: List[str] = []
|
55 |
-
|
56 |
-
for query in queries:
|
57 |
-
query = f"Question: {query}"
|
58 |
-
query += suffix # add suffix (pad tokens)
|
59 |
-
texts_query.append(query)
|
60 |
-
|
61 |
-
batch_query = self.tokenizer(
|
62 |
-
#images=[self.mock_image] * len(texts_query),
|
63 |
-
text=texts_query,
|
64 |
-
return_tensors="pt",
|
65 |
-
padding="longest",
|
66 |
-
max_length= max_length + self.image_seq_length,
|
67 |
-
)
|
68 |
-
|
69 |
-
return batch_query
|
70 |
-
|
71 |
-
def score(
|
72 |
-
self,
|
73 |
-
qs: List[torch.Tensor],
|
74 |
-
ps: List[torch.Tensor],
|
75 |
-
device: Optional[Union[str, torch.device]] = None,
|
76 |
-
**kwargs,
|
77 |
-
) -> torch.Tensor:
|
78 |
-
"""
|
79 |
-
Compute the MaxSim score (ColBERT-like) for the given multi-vector query and passage embeddings.
|
80 |
-
"""
|
81 |
return self.score_multi_vector(qs, ps, device=device, **kwargs)
|
|
|
1 |
+
from typing import List, Optional, Union
|
2 |
+
|
3 |
+
import torch
|
4 |
+
from PIL import Image
|
5 |
+
from transformers import BatchFeature
|
6 |
+
|
7 |
+
from processing_florence2 import Florence2Processor
|
8 |
+
|
9 |
+
from colpali_engine.utils.processing_utils import BaseVisualRetrieverProcessor
|
10 |
+
|
11 |
+
|
12 |
+
class ColFlorProcessor(BaseVisualRetrieverProcessor, Florence2Processor):
|
13 |
+
"""
|
14 |
+
Processor for ColPali.
|
15 |
+
"""
|
16 |
+
|
17 |
+
def __init__(self, *args, **kwargs):
|
18 |
+
super().__init__(*args, **kwargs)
|
19 |
+
self.mock_image = Image.new("RGB", (16, 16), color="black")
|
20 |
+
|
21 |
+
def process_images(
|
22 |
+
self,
|
23 |
+
images: List[Image.Image],
|
24 |
+
) -> BatchFeature:
|
25 |
+
"""
|
26 |
+
Process images for ColFlor2.
|
27 |
+
"""
|
28 |
+
texts_doc = ["<OCR>"] * len(images)
|
29 |
+
images = [image.convert("RGB") for image in images]
|
30 |
+
|
31 |
+
batch_doc = self(
|
32 |
+
text=texts_doc,
|
33 |
+
images=images,
|
34 |
+
return_tensors="pt",
|
35 |
+
padding="longest",
|
36 |
+
)
|
37 |
+
|
38 |
+
new_part = torch.ones((batch_doc['attention_mask'].size()[0], 577)).to(batch_doc['attention_mask'].device)
|
39 |
+
batch_doc['full_attention_mask'] = torch.cat([new_part, batch_doc['attention_mask']], dim=1)
|
40 |
+
|
41 |
+
return batch_doc
|
42 |
+
|
43 |
+
def process_queries(
|
44 |
+
self,
|
45 |
+
queries: List[str],
|
46 |
+
max_length: int = 50,
|
47 |
+
suffix: Optional[str] = None,
|
48 |
+
) -> BatchFeature:
|
49 |
+
"""
|
50 |
+
Process queries for ColFlor2.
|
51 |
+
"""
|
52 |
+
if suffix is None:
|
53 |
+
suffix = "<pad>" * 10
|
54 |
+
texts_query: List[str] = []
|
55 |
+
|
56 |
+
for query in queries:
|
57 |
+
query = f"Question: {query}"
|
58 |
+
query += suffix # add suffix (pad tokens)
|
59 |
+
texts_query.append(query)
|
60 |
+
|
61 |
+
batch_query = self.tokenizer(
|
62 |
+
#images=[self.mock_image] * len(texts_query),
|
63 |
+
text=texts_query,
|
64 |
+
return_tensors="pt",
|
65 |
+
padding="longest",
|
66 |
+
max_length= max_length + self.image_seq_length,
|
67 |
+
)
|
68 |
+
|
69 |
+
return batch_query
|
70 |
+
|
71 |
+
def score(
|
72 |
+
self,
|
73 |
+
qs: List[torch.Tensor],
|
74 |
+
ps: List[torch.Tensor],
|
75 |
+
device: Optional[Union[str, torch.device]] = None,
|
76 |
+
**kwargs,
|
77 |
+
) -> torch.Tensor:
|
78 |
+
"""
|
79 |
+
Compute the MaxSim score (ColBERT-like) for the given multi-vector query and passage embeddings.
|
80 |
+
"""
|
81 |
return self.score_multi_vector(qs, ps, device=device, **kwargs)
|