edugp's picture
Support png and jpg
7b3d1d9
raw
history blame
1.69 kB
import os
import sys
import streamlit as st
import transformers
from huggingface_hub import snapshot_download
from transformers import AutoTokenizer
LOCAL_PATH = snapshot_download("flax-community/clip-spanish")
sys.path.append(LOCAL_PATH)
from modeling_hybrid_clip import FlaxHybridCLIP
from test_on_image import run_inference
def save_file_to_disk(uplaoded_file):
temp_file = os.path.join("/tmp", uplaoded_file.name)
with open(temp_file, "wb") as f:
f.write(uploaded_file.getbuffer())
return temp_file
@st.cache(
hash_funcs={
transformers.models.bert.tokenization_bert_fast.BertTokenizerFast: id,
FlaxHybridCLIP: id,
}
)
def load_tokenizer_and_model():
# load the saved model
tokenizer = AutoTokenizer.from_pretrained("dccuchile/bert-base-spanish-wwm-cased")
model = FlaxHybridCLIP.from_pretrained(LOCAL_PATH)
return tokenizer, model
tokenizer, model = load_tokenizer_and_model()
st.title("Caption Scoring")
uploaded_file = st.file_uploader("Choose an image...", type=["png", "jpg"])
text_input = st.text_input("Type a caption")
if uploaded_file is not None and text_input:
local_image_path = None
try:
local_image_path = save_file_to_disk(uploaded_file)
score = run_inference(local_image_path, text_input, model, tokenizer).tolist()
st.image(
uploaded_file,
caption=text_input,
width=None,
use_column_width=None,
clamp=False,
channels="RGB",
output_format="auto",
)
st.write(f"## Score: {score:.2f}")
finally:
if local_image_path:
os.remove(local_image_path)