Delete streamlit_app.py

streamlit_app.py

@@ -1,101 +0,0 @@
-# streamlit_app.py
-import streamlit as st
-from PIL import Image
-import torch
-import torchvision.transforms as T
-from transformers import ViTModel, T5ForConditionalGeneration, T5Tokenizer
-
-# ————— Model & tokenizer settings ——————————
-HF_MODEL_ID = "your-username/iu-chest-report-model"  # ← replace with your model repo
-
-@st.cache_resource(show_spinner=False)
-def load_models(model_id):
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    vit = ViTModel.from_pretrained("google/vit-base-patch16-224").to(device)
-    t5  = T5ForConditionalGeneration.from_pretrained(model_id).to(device)
-    tok = T5Tokenizer.from_pretrained(model_id)
-    return device, vit, t5, tok
-
-device, vit, t5, tokenizer = load_models(HF_MODEL_ID)
-
-# ————— Transforms ——————————————————————————
-transform = T.Compose([
-    T.Resize((224,224)),
-    T.ToTensor(),
-    T.Normalize(mean=0.5, std=0.5),
-])
-
-# ————— Streamlit layout ——————————————————————
-st.set_page_config(page_title="Radiology Report Analysis", layout="wide")
-st.markdown("<h1 style='text-align:center;'>Radiology Report Analysis</h1>", unsafe_allow_html=True)
-st.markdown("<p style='text-align:center;'>Upload a chest X-ray (JPG or PNG) to let our AI analyze and provide a diagnosis.</p>", unsafe_allow_html=True)
-
-# Session state to control flow
-if "stage" not in st.session_state:
-    st.session_state.stage = "upload"
-
-# 1) Upload stage
-if st.session_state.stage == "upload":
-    uploader = st.file_uploader("", type=["png","jpg","jpeg"], label_visibility="collapsed")
-    gen_button = st.button("Generate Report", disabled=(uploader is None))
-    
-    if uploader:
-        st.image(uploader, width=400, caption=f"{uploader.name} ({uploader.size/1e6:.2f} MB)")
-    if gen_button:
-        st.session_state.uploaded = uploader
-        st.session_state.stage = "report"
-
-# 2) Report stage
-elif st.session_state.stage == "report":
-    img_file = st.session_state.uploaded
-    img = Image.open(img_file).convert("RGB")
-    
-    # Run inference once
-    with st.spinner("Analyzing X-ray…"):
-        # 1) preprocess
-        x = transform(img).unsqueeze(0).to(device)
-        # 2) vision features
-        vfeat = vit(pixel_values=x).pooler_output      # [1,H]
-        # 3) prepare a dummy prompt - adapt if your model expects something else
-        prompt = "report:"
-        enc = tokenizer(prompt, return_tensors="pt").to(device)
-        # 4) generate
-        gen_ids = t5.generate(
-            input_ids=enc.input_ids,
-            attention_mask=enc.attention_mask,
-            max_length=64,
-            num_beams=4,
-            no_repeat_ngram_size=2,
-            early_stopping=True,
-            length_penalty=1.2,
-        )
-        diagnosis = tokenizer.decode(gen_ids[0], skip_special_tokens=True)
-        confidence = 94  # or compute a proxy if you have one
-
-    # Display results
-    col1, col2 = st.columns([1,1])
-    with col1:
-        st.subheader("Your Uploaded X-ray")
-        st.image(img, use_column_width=True)
-        st.markdown(f"**File:** {img_file.name}  \n**Size:** {img_file.size/1e6:.2f} MB")
-    with col2:
-        st.subheader("AI Diagnosis & Report")
-        st.markdown(
-            f"<div style='background:#e0f7fa;padding:12px;border-radius:6px;'>"
-            f"<strong>Primary Diagnosis</strong><br>{diagnosis}</div>",
-            unsafe_allow_html=True
-        )
-        st.markdown(f"**Confidence:** {confidence}%")
-
-    if st.button("⬅️ Upload Another"):
-        # reset
-        st.session_state.stage = "upload"
-        del st.session_state.uploaded
-
-# ————— Footer ——————————————————————————————
-st.markdown("""
-<hr style='margin:2em 0;'>
-<p style='font-size:0.8em;color:gray;text-align:center;'>
-  This app runs a fine-tuned Vision→Report model from Hugging Face Space.
-</p>
-""", unsafe_allow_html=True)