test gridio

README.md

@@ -3,7 +3,7 @@ title: Medical diagnosis evaluation via MedCLIP
 emoji: 👩‍⚕️
 colorFrom: blue
 colorTo: indigo
-sdk: streamlit
+sdk: gradio
 app_file: app.py
 pinned: false
 ---

app.py

@@ -1,9 +1,9 @@
-import os
-import sys
+import sys 
+
+import gradio as gr
 import jax
-import streamlit as st
-import transformers
 from huggingface_hub import snapshot_download
+from PIL import Image
 from transformers import AutoTokenizer
 import torch
 from torchvision.io import ImageReadMode, read_image
@@ -15,7 +15,6 @@ sys.path.append(LOCAL_PATH)
 from src.modeling_medclip import FlaxMedCLIP
 from run_medclip import Transform
 
-
 def prepare_image(image_path, model):
     image = read_image(image_path, mode=ImageReadMode.RGB)
     preprocess = Transform(model.config.vision_config.image_size)
@@ -28,17 +27,11 @@ def prepare_text(text, tokenizer):
     return tokenizer(text, return_tensors="np")
 
 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())
+    temp_file = "/tmp/image.jpeg"
+    im = Image.fromarray(uplaoded_file)
+    im.save(temp_file)
     return temp_file
-@st.cache(
-    hash_funcs={
-        transformers.models.bert.tokenization_bert_fast.BertTokenizerFast: id,
-        FlaxMedCLIP: id,
-    },
-    show_spinner=False
-)
+
 def load_tokenizer_and_model():
     # load the saved model
     tokenizer = AutoTokenizer.from_pretrained("allenai/scibert_scivocab_uncased")
@@ -60,24 +53,16 @@ def run_inference(image_path, text, model, tokenizer):
     return score
 
 tokenizer, model = load_tokenizer_and_model()
-st.title("Diagnosis Scoring")
-uploaded_file = st.file_uploader("Choose a Chest x-ray...", type=["png", "jpg"])
-text_input = st.text_input("Type the doctor diagnosis")
-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)
+
+def score_image_caption_pair(uploaded_file, text_input):
+    local_image_path = save_file_to_disk(uploaded_file)
+    score = run_inference(
+        local_image_path, text_input, model, tokenizer).tolist()
+    return {"Score": score},  "{:.2f}".format(score)
+
+
+image = gr.inputs.Image(shape=(299, 299))
+iface = gr.Interface(
+    fn=score_image_caption_pair, inputs=[image, "text"], outputs=["label", "text"]
+)
+iface.launch()

requirements.txt

@@ -1,7 +1,7 @@
 flax==0.3.4
 huggingface-hub==0.0.12
 jax==0.2.17
-streamlit==0.84.1
+gradio==2.2.2
 torch==1.9.0
 torchvision==0.10.0
 transformers==4.8.2