Upload 3 files

models/llm.py

@@ -0,0 +1,126 @@
+
+from transformers import pipeline
+
+class StyleSavvy:
+    def __init__(
+        self,
+        model_name: str = "google/flan-t5-large",
+        device: int   = -1,     # -1 = CPU, or GPU index
+        max_length: int = 150,
+    ):
+        # A local instruction-tuned T5 model
+        self.pipe = pipeline(
+            "text2text-generation",
+            model=model_name,
+            tokenizer=model_name,
+            device=device,
+        )
+        self.max_length = max_length
+    # TODO: Modification: Add more prompts to the advise function
+    # to make it more specific to the user's needs.
+    # The function now takes in the user's body type, face shape, and occasion
+    # and generates style tips accordingly.
+
+    def advise(self, items, body_type, face_shape, occasion):
+        prompt = (
+            f"The user is {body_type}-shaped with a {face_shape} face, "
+            f"attending a {occasion}. They are wearing: "
+            + ", ".join(i["label"] for i in items)
+            + ".\n\nPlease list 5 concise style tips as bullet points:"
+        )
+
+        # Generate with supported args only
+        result = self.pipe(
+            prompt,
+            max_length=self.max_length,
+            num_beams=4,
+            early_stopping=True,
+            do_sample=False
+        )[0]["generated_text"].strip()
+        
+        return result
+
+
+
+
+# import torch
+
+# # models/llm.py
+
+# # models/llm.py
+
+# import os
+# from typing import List
+# from transformers import pipeline, Pipeline
+
+# # Force CPU modes (avoid any MPS/CUDA issues on macOS)
+# os.environ["CUDA_VISIBLE_DEVICES"] = ""
+# os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "0"
+
+# class StyleSavvy:
+#     def __init__(
+#         self,
+#         model_name: str = "openlm-research/open_llama_3b_v2",
+#         device: int   = -1,        # -1 = CPU
+#         max_new_tokens: int = 100,
+#         temperature: float   = 0.7,
+#         top_p: float         = 0.9,
+#     ):
+#         """
+#         Uses OpenLLaMA-3B-v2 (≈3B params) for fast, local inference.
+#         """
+#         # Setup a causal text-generation pipeline
+#         self.pipe: Pipeline = pipeline(
+#             "text-generation",
+#             model=model_name,
+#             tokenizer=model_name,
+#             device=device,
+#         )
+#         # GPT‐style models need a pad token to avoid warnings
+#         if self.pipe.tokenizer.pad_token_id is None:
+#             self.pipe.tokenizer.pad_token = self.pipe.tokenizer.eos_token
+        
+#         self.max_new_tokens = max_new_tokens
+#         self.temperature    = temperature
+#         self.top_p          = top_p
+
+#     def advise(
+#         self,
+#         items: List[str],
+#         body_type: str,
+#         face_shape: str,
+#         occasion: str
+#     ) -> List[str]:
+#         """
+#         Builds a strict instruction prompt and returns exactly five "- " bullets.
+#         """
+#         labels = ", ".join(items) if items else "an outfit"
+#         prompt = (
+#             "You are a professional fashion consultant.\n"
+#             f"The user is {body_type}-shaped with a {face_shape} face, attending {occasion}.\n"
+#             f"They are wearing: {labels}.\n\n"
+#             "Please provide exactly five concise style tips, each on its own line, "
+#             "and starting with \"- \". No extra text."
+#         )
+
+#         # Generate
+#         output = self.pipe(
+#             prompt,
+#             max_new_tokens=self.max_new_tokens,
+#             do_sample=True,
+#             temperature=self.temperature,
+#             top_p=self.top_p,
+#             return_full_text=False,
+#         )[0]["generated_text"]
+
+#         # Extract bullets
+#         tips = [ln.strip() for ln in output.splitlines() if ln.strip().startswith("- ")]
+#         # Fallback: split on sentences if fewer than 5 bullets
+#         if len(tips) < 5:
+#             candidates = [s.strip() for s in output.replace("\n"," ").split(".") if s.strip()]
+#             tips = [f"- {candidates[i]}" for i in range(min(5, len(candidates)))]
+        
+#         return tips[:5]
+
+
+

models/test_llm.py

@@ -0,0 +1,162 @@
+# # test_llm.py
+# """
+# Test harness for StyleSavvy LLM prompts.
+# Defines multiple prompt templates and evaluates the generated outputs,
+# checking for the expected number of bullet-point style tips.
+# """
+# from models.llm import StyleSavvy
+
+# # Variant prompt templates with placeholders
+# PROMPT_TEMPLATES = {
+#     "occasion_driven": (
+#         "You are an expert fashion stylist. A client is preparing for {occasion}. "
+#         "They have a {body_type}-shaped body and a {face_shape} face. They’re currently wearing: {items}. "
+#         "Give 3 to 5 *distinct* style tips focused on making them look their best at the event. "
+#         "Make the suggestions relevant to the setting, weather, and formality of the occasion. "
+#         "Avoid repeating any advice."
+#     ),
+
+#     "function_based": (
+#         "You're advising someone with a {body_type} build and {face_shape} face. "
+#         "They're attending a {occasion} and are wearing {items}. "
+#         "Suggest 3–5 concise fashion improvements or enhancements. "
+#         "Each suggestion should be unique and tailored to the event. "
+#         "Include practical choices for color, layering, accessories, or footwear. "
+#         "Avoid repeating words or phrases."
+#     ),
+
+#     "intent_style": (
+#         "Act as a high-end personal stylist. Your client has a {body_type} body shape and a {face_shape} face. "
+#         "They're going to a {occasion} and are wearing {items}. "
+#         "Write 3 to 5 brief but powerful styling suggestions to elevate their look. "
+#         "Focus on intent—what feeling or impression each style choice creates for the event."
+#     ),
+# }
+
+
+# # Test parameters
+# BODY_TYPE = "Slim"
+# FACE_SHAPE = "Round"
+# OCCASION = "Rooftop Evening Party"
+# ITEMS = ["shirt", "jeans", "jacket","shoes"]
+
+# if __name__ == "__main__":
+#     advisor = StyleSavvy()
+
+#     for name, template in PROMPT_TEMPLATES.items():
+#         # Build prompt by replacing placeholders
+#         prompt = template.format(
+#             body_type=BODY_TYPE,
+#             face_shape=FACE_SHAPE,
+#             occasion=OCCASION,
+#             items=", ".join(ITEMS)
+#         )
+#         print(f"=== Testing template: {name} ===")
+#         print("Prompt:")
+#         print(prompt)
+
+#         # Generate output (use only supported args)
+#         result = advisor.pipe(
+#             prompt,
+#             max_length=advisor.max_length,
+#             early_stopping=True,
+#             do_sample=False
+#         )[0]["generated_text"].strip()
+
+#         print("Generated output:")
+#         print(result)
+
+#         # Extract bullet lines
+#         bullets = [ln for ln in result.splitlines() if ln.strip().startswith("- ")]
+#         print(f"Number of bullets detected: {len(bullets)}")
+#         for i, b in enumerate(bullets, start=1):
+#             print(f" {i}. {b}")
+#         print("" + "-"*40)
+
+
+# test_llm.py
+"""
+Test harness for StyleSavvy LLM prompts.
+Evaluates multiple prompt templates and parses the generated outputs into distinct tips.
+"""
+
+from models.llm import StyleSavvy
+
+# Variant prompt templates with placeholders
+PROMPT_TEMPLATES = {
+    "direct_instruction": (
+        "You are a professional fashion stylist. A client with a {body_type} body shape "
+        "and {face_shape} face is preparing for a {occasion}. They are currently wearing {items}. "
+        "Give exactly five different styling tips to improve their outfit. "
+        "Each tip should be concise, actionable, and relevant to the event. Start each tip on a new line."
+    ),
+    "category_expansion": (
+        "As a high-end fashion advisor, provide five styling tips for a {body_type}-shaped person "
+        "with a {face_shape} face attending a {occasion}. They are currently wearing {items}. "
+        "Offer one tip for each of the following categories: silhouette, color, accessories, footwear, and layering. "
+        "Each tip must be brief, specific, and clearly separated by a line break."
+    ),
+    "event_aesthetic": (
+        "Imagine you're curating a perfect outfit for a {body_type}-shaped individual with a {face_shape} face "
+        "attending {occasion}. They are wearing {items}. Suggest 5 ways to enhance their style, focusing on event-appropriate aesthetics. "
+        "Write each tip as a separate sentence on a new line. Do not repeat advice or themes."
+    ),
+    "fashion_editor": (
+        "As a fashion editor writing for a style magazine, outline five unique styling tips for a {body_type}-shaped reader "
+        "with a {face_shape} face who is attending {occasion}. They currently wear {items}. "
+        "Each recommendation should reflect expertise, relevance to the occasion, and a unique style element. "
+        "Deliver all five tips in a list format, starting each on a new line."
+    ),
+    "influencer_style": (
+        "You’re an influencer known for your sharp styling advice. One of your followers has a {body_type} body and "
+        "{face_shape} face, and they're attending {occasion}. They’ve sent you a photo wearing {items}. "
+        "Reply with exactly five snappy, modern style tips they can use to upgrade their outfit for the event. "
+        "Make sure each tip is short, non-repetitive, and on its own line."
+    ),
+}
+
+
+# Test parameters
+BODY_TYPE = "Slim"
+FACE_SHAPE = "Round"
+OCCASION = "Rooftop Evening Party"
+ITEMS = ["jeans", "jacket", "shoes"]
+
+if __name__ == "__main__":
+    advisor = StyleSavvy()
+
+    for name, template in PROMPT_TEMPLATES.items():
+        print(f"=== Testing template: {name} ===")
+
+        # Build prompt
+        prompt = template.format(
+            body_type=BODY_TYPE,
+            face_shape=FACE_SHAPE,
+            occasion=OCCASION,
+            items=", ".join(ITEMS)
+        )
+        print("Prompt:\n" + prompt)
+
+        # Generate response
+        result = advisor.pipe(
+            prompt,
+            max_length=advisor.max_length,
+            early_stopping=True,
+            num_beams=4,
+            no_repeat_ngram_size=3,  
+            do_sample=False)[0]["generated_text"].strip()
+
+        print("\nRaw generated output:\n" + result)
+
+        # Parse into tips (bullets or sentence)
+        lines = result.splitlines()
+        tips = [ln.strip("-*0123456789. ").strip() for ln in lines if ln.strip()]
+        if len(tips) < 3:
+            # fallback to sentence split
+            tips = [p.strip() for p in result.split(".") if p.strip()]
+        tips = list(dict.fromkeys(tips))  # remove duplicates
+
+        print(f"\n💡 Parsed {len(tips)} style tips:")
+        for i, tip in enumerate(tips[:5], 1):
+            print(f"{i}. {tip}")
+        print("-" * 40)

models/vision.py

@@ -0,0 +1,50 @@
+
+# models/vision.py -- Working
+
+from transformers import pipeline
+from PIL import Image
+
+class VisionModel:
+    def __init__(
+        self,
+        model_name: str = "valentinafeve/yolos-fashionpedia",
+        threshold: float = 0.7
+    ):
+        self.pipe = pipeline("object-detection", model=model_name)
+        self.threshold = threshold
+
+    def detect(self, image: Image.Image):
+        # 1) Ensure RGB
+        if image.mode != "RGB":
+            image = image.convert("RGB")
+
+        # 2) Run detection
+        results = self.pipe(image)
+
+        # 3) Process & filter
+        processed = []
+        for r in results:
+            score = float(r["score"])
+            if score < self.threshold:
+                continue
+
+            # r["box"] is a dict: {"xmin":..., "ymin":..., "xmax":..., "ymax":...}
+            box = r["box"]
+            coords = [
+                float(box["xmin"]),
+                float(box["ymin"]),
+                float(box["xmax"]),
+                float(box["ymax"]),
+            ]
+
+            processed.append({
+                "label": r["label"],
+                "score": score,
+                "box": coords
+            })
+
+        return processed
+
+
+
+