Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -4,10 +4,32 @@ import PIL.Image
|
|
4 |
import skimage.io as io
|
5 |
import streamlit as st
|
6 |
from transformers import GPT2Tokenizer, GPT2LMHeadModel, AdamW, get_linear_schedule_with_warmup
|
|
|
7 |
from model import generate2,ClipCaptionModel
|
8 |
from engine import inference
|
9 |
|
10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
device = "cpu"
|
13 |
clip_model, preprocess = clip.load("ViT-B/32", device=device, jit=False)
|
@@ -35,7 +57,7 @@ def ui():
|
|
35 |
pil_image = PIL.Image.fromarray(image)
|
36 |
image = preprocess(pil_image).unsqueeze(0).to(device)
|
37 |
|
38 |
-
option = st.selectbox('Please select the Model',('Model', 'COCO Model','PreTrained Model'))
|
39 |
|
40 |
if option=='Model':
|
41 |
with torch.no_grad():
|
@@ -60,6 +82,12 @@ def ui():
|
|
60 |
st.image(uploaded_file, width = 500, channels = 'RGB')
|
61 |
st.markdown("**PREDICTION:** " + out)
|
62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
|
64 |
if __name__ == '__main__':
|
65 |
ui()
|
|
|
4 |
import skimage.io as io
|
5 |
import streamlit as st
|
6 |
from transformers import GPT2Tokenizer, GPT2LMHeadModel, AdamW, get_linear_schedule_with_warmup
|
7 |
+
from transformers import GPT2TokenizerFast, ViTImageProcessor, VisionEncoderDecoderModel
|
8 |
from model import generate2,ClipCaptionModel
|
9 |
from engine import inference
|
10 |
|
11 |
+
|
12 |
+
model_trained = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
|
13 |
+
model_trained.load_state_dict(torch.load('model_trained.pth',map_location=torch.device('cpu')))
|
14 |
+
image_processor = ViTImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
|
15 |
+
tokenizer = GPT2TokenizerFast.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
|
16 |
+
|
17 |
+
def show_n_generate(img, greedy = True, model = model_raw):
|
18 |
+
image = Image.open(img)
|
19 |
+
pixel_values = image_processor(image, return_tensors ="pt").pixel_values
|
20 |
+
plt.imshow(np.asarray(image))
|
21 |
+
plt.show()
|
22 |
+
|
23 |
+
if greedy:
|
24 |
+
generated_ids = model.generate(pixel_values, max_new_tokens = 30)
|
25 |
+
else:
|
26 |
+
generated_ids = model.generate(
|
27 |
+
pixel_values,
|
28 |
+
do_sample=True,
|
29 |
+
max_new_tokens = 30,
|
30 |
+
top_k=5)
|
31 |
+
generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
32 |
+
returned generated_text
|
33 |
|
34 |
device = "cpu"
|
35 |
clip_model, preprocess = clip.load("ViT-B/32", device=device, jit=False)
|
|
|
57 |
pil_image = PIL.Image.fromarray(image)
|
58 |
image = preprocess(pil_image).unsqueeze(0).to(device)
|
59 |
|
60 |
+
option = st.selectbox('Please select the Model',('Model', 'COCO Model','PreTrained Model','Fine Tuned Model'))
|
61 |
|
62 |
if option=='Model':
|
63 |
with torch.no_grad():
|
|
|
82 |
st.image(uploaded_file, width = 500, channels = 'RGB')
|
83 |
st.markdown("**PREDICTION:** " + out)
|
84 |
|
85 |
+
elif option=='Fine Tuned Model':
|
86 |
+
out=show_n_generate(uploaded_file, greedy = False, model = model_trained)
|
87 |
+
st.image(uploaded_file, width = 500, channels = 'RGB')
|
88 |
+
st.markdown("**PREDICTION:** " + out)
|
89 |
+
|
90 |
+
|
91 |
|
92 |
if __name__ == '__main__':
|
93 |
ui()
|