Add execution logic

app.py

@@ -4,7 +4,9 @@ import torch
 import logging
 import json
 import pickle
+from PIL import Image
 import gradio as gr
+from scipy.special import softmax
 
 
 logger = logging.getLogger("basebody")
@@ -14,7 +16,6 @@ TEXT_PROMPTS_FILE_NAME = "text_prompts.json"
 LOGISTIC_REGRESSION_MODEL_FILE_NAME = "logistic_regression_l1_oct_2.pkl"
 
 
-
 clip_model, preprocess = clip.load(CLIP_MODEL_NAME, device="cpu")
 
 with open(
@@ -44,9 +45,29 @@ with torch.no_grad():
 
 
 def predict_fn(input_img):
-    print(type(input_img))
-    print(input_img)
-    return "Hello " + "!"
+    input_img = Image.fromarray(input_img.astype("uint8"), "RGB")
+    image = preprocess(
+        input_img
+    ).unsqueeze(0)
+    with torch.no_grad():
+        image_features = clip_model.encode_image(image)
+        cosine_simlarities = softmax(
+            (all_text_features @ image_features.cpu().T)
+            .squeeze()
+            .reshape(len(text_prompts), 2, -1),
+            axis=1,
+        )[:, 0, :]
+        # logger.info(f"cosine_simlarities shape: {cosine_simlarities.shape}")
+        logger.info(f"cosine_simlarities: {cosine_simlarities}")
+    probabilities = lr_model.predict_proba(
+        cosine_simlarities.reshape(1, -1)
+    )
+    logger.info(f"probabilities: {probabilities}")
+    decision_json = json.dumps(
+        {"is_base_body": float(probabilities[0][1])}
+    ).encode("utf-8")
+    logger.info(f"decision_json: {decision_json}")
+    return decision_json
 
 
 iface = gr.Interface(