Spaces:
Runtime error
Runtime error
baixintech_zhangyiming_prod
commited on
Commit
·
141d444
1
Parent(s):
3d7e2fe
init
Browse files- app.py +30 -0
- requirements.txt +3 -0
app.py
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import gradio.components as grc
|
| 3 |
+
import torch
|
| 4 |
+
from PIL import Image
|
| 5 |
+
|
| 6 |
+
from lavis.models import load_model_and_preprocess
|
| 7 |
+
from lavis.processors import load_processor
|
| 8 |
+
|
| 9 |
+
raw_image = Image.open("merlion.png").convert("RGB")
|
| 10 |
+
|
| 11 |
+
# setup device to use
|
| 12 |
+
device = torch.device("cuda") if torch.cuda.is_available() else "cpu"
|
| 13 |
+
caption = "merlion in Singapore"
|
| 14 |
+
|
| 15 |
+
model, vis_processors, text_processors = load_model_and_preprocess("blip2_image_text_matching", "pretrain", device=device, is_eval=True)
|
| 16 |
+
|
| 17 |
+
def predict(raw_image, caption):
|
| 18 |
+
raw_image = raw_image.convert("RGB")
|
| 19 |
+
img = vis_processors["eval"](raw_image).unsqueeze(0).to(device)
|
| 20 |
+
txt = text_processors["eval"](caption)
|
| 21 |
+
itm_output = model({"image": img, "text_input": txt}, match_head="itm")
|
| 22 |
+
itm_scores = torch.nn.functional.softmax(itm_output, dim=1)
|
| 23 |
+
itm_score = itm_scores[:, 1].item()
|
| 24 |
+
itc_score = model({"image": img, "text_input": txt}, match_head='itc')
|
| 25 |
+
return '%.3f' % itm_score, '%.4f' % itc_score
|
| 26 |
+
|
| 27 |
+
app = gr.Interface(fn=predict, inputs=[grc.Image(type="pil"), grc.Textbox()], outputs=[grc.Text(label="itm score"), grc.Text(label="itc score")])
|
| 28 |
+
app.launch()
|
| 29 |
+
|
| 30 |
+
|
requirements.txt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
salesforce-lavis
|
| 2 |
+
pillow
|
| 3 |
+
torch
|