import io import os import ast import json import base64 import spaces import requests import numpy as np import gradio as gr from PIL import Image from io import BytesIO import face_recognition from turtle import title from openai import OpenAI from collections import defaultdict from typing import List, Optional, Set, Dict, Any from transformers import pipeline import urllib.request from transformers import YolosImageProcessor, YolosForObjectDetection import torch import matplotlib.pyplot as plt from torchvision.transforms import ToTensor, ToPILImage client = OpenAI() pipe = pipeline("zero-shot-image-classification", model="patrickjohncyh/fashion-clip") color_file_path = 'color_config.json' attributes_file_path = 'attributes_config.json' OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") # Open and read the COLOR JSON file with open(color_file_path, 'r') as file: color_data = json.load(file) # Open and read the ATTRIBUTES JSON file with open(attributes_file_path, 'r') as file: attributes_data = json.load(file) COLOURS_DICT = color_data['color_mapping'] ATTRIBUTES_DICT = attributes_data['attribute_mapping'] DETAILS_THRESHOLD = 0.4 # This is how high the total score of an additional detail attribute should be for it to be included (number scales up linearly with more images) OPENAI_USER_PROMPT = "I have a product called 'Round Neck Knitted Ribbed Crop Top with a Neck Detail'. The description reads: 'Introducing the Round Neck Knitted Ribbed Crop Top with a Neck Detail. This sleeveless top is a must-have addition to your wardrobe. Its round neck and ribbed design give it a modern and stylish look, while the neck detail adds a unique touch. Made from high-quality materials, this crop top is comfortable to wear all day long. The solid white color makes it easy to pair with any outfit, making it a versatile piece for any occasion. Whether you're dressing up for a night out or keeping it casual, this crop top is sure to become one of your favorite go-to staples. Upgrade your style with this trendy and chic top today!'. Arabic naming: 'بلوزة قصيرة محبوكة ومضلعة بياقة دائرية وتفاصيل رقبة'. Arabic description: 'نقدم لك بلوزة قصيرة محبوكة ومضلعة برقبة دائرية مع تفاصيل الرقبة. يعد هذا القميص بلا أكمام إضافة لا غنى عنها لخزانة ملابسك. تمنحه رقبته المستديرة وتصميمه المضلع مظهرًا عصريًا وأنيقًا، بينما تضيف تفاصيل الرقبة لمسة فريدة. مصنوع من مواد عالية الجودة، هذا القميص القصير مريح للارتداء طوال اليوم. اللون الأبيض الصلب يجعل من السهل ارتداؤه مع أي ملابس، مما يجعله قطعة متعددة الاستخدامات لأي مناسبة. سواء كنت ترتدي ملابس ليلية أو تبقيها غير رسمية، فمن المؤكد أن هذه السترة القصيرة ستصبح واحدة من القطع الأساسية المفضلة لديك. قم بترقية أسلوبك مع هذا القميص العصري والأنيق اليوم!'\n The attributes are:\n\t• Category: 'top > croptop'\n\t• Material: 'polyamide'\n\t• Target Audience: women\n\t• Fabric Style: Knitted\n\t• Pattern: Solid\n\t• Colar Type: Round Neck\n\t• Sleeve Length: Sleeveless\n\t• Details: Ribbed Neck Detail \n\t• Tags: 'neutral, staples, casual, springsummer, bestseller, summer_light_palette, coastal_charm, allnewin' (do not use any seasonal tags or other discount info nothing with '_' or color)\n \n\nI have a product called 'Crew Neck Crop Top with a V-Shaped Front'. The description reads: 'Step up your casual style with this Crew Neck Crop Top with a V-Shaped Front. Crafted from high-quality materials, this sleeveless top features a flattering v-shaped front design that adds a touch of sophistication to any outfit. The Sacramento Green color adds a pop of freshness and pairs perfectly with neutral tones. Whether you're lounging at home or running errands, this crop top is a staple in your wardrobe. Its versatile design makes it easy to dress up or down for any occasion. Elevate your loungewear game and add this Crew Neck Crop Top to your collection today.'. Arabic naming: 'بلوزة قصيرة بياقة دائرية ومقدمة على شكل حرف V'. Arabic description: 'ارتقي بأسلوبك الكاجوال مع هذا القميص القصير ذو الرقبة الدائرية والمقدمة على شكل حرف V. مصنوع من مواد عالية الجودة، يتميز هذا القميص بدون أكمام بتصميم أمامي جذاب على شكل حرف V يضيف لمسة من الرقي إلى أي ملابس. يضيف لون Sacramento Green لمسة من النضارة ويمتزج بشكل مثالي مع الألوان المحايدة. سواء كنت تسترخي في المنزل أو تقوم بمهمات، فإن هذا القميص القصير هو قطعة أساسية في خزانة ملابسك. تصميمه متعدد الاستخدامات يجعل من السهل ارتداؤه لأعلى أو لأسفل في أي مناسبة. ارفع مستوى ملابسك المريحة وأضف هذا القميص ذو الرقبة الدائرية إلى مجموعتك اليوم.'\n The attributes are:\n\t• Category: 'top > croptop'\n\t• Material: 'cotton, polyester'\n\t• Target Audience: women\n\t• Fabric Style: Woven\n\t• Pattern: Solid\n\t• Colar Type: Crew Neck\n\t• Sleeve Length: Sleeveless\n\t• Details: V-Shaped Front \n\t• Tags: 'neutral, casual, staples, loungewear, bestseller' (do not use any seasonal tags or other discount info nothing with '_' or color)\n \n\nI have a product called 'Corduroy Polo Collar Buttoned Pocket Jacket'. The description reads: 'Step up your fashion game with this Corduroy Polo Collar Buttoned Pocket Jacket. Made from high-quality corduroy fabric, this jacket is both stylish and comfortable. The unique polo collar adds a touch of sophistication, while the buttoned pockets provide a practical yet fashionable twist. Whether you're heading to a casual outing or a night on the town, this jacket is sure to turn heads. Its versatile design makes it easy to pair with any outfit, giving you endless options for creating trendy looks. Upgrade your wardrobe with this Hiccup fashion statement and be the envy of everyone around you.'. Arabic naming: 'سترة سروال قصير بياقة بولو وجيب بأزرار'. Arabic description: 'ارتقِ بأسلوبك في عالم الموضة مع سترة الجيب ذات الأزرار بياقة بولو من سروال قصير. مصنوع من قماش كوردروي عالي الجودة، هذا الجاكيت أنيق ومريح. تضيف ياقة البولو الفريدة لمسة من الرقي، بينما توفر الجيوب ذات الأزرار لمسة عملية وعصرية. سواء كنت متجهًا إلى نزهة غير رسمية أو ليلة في المدينة، فمن المؤكد أن هذه السترة ستلفت الأنظار. تصميمه متعدد الاستخدامات يجعل من السهل ارتداؤه مع أي ملابس، مما يمنحك خيارات لا حصر لها للحصول على إطلالات عصرية. قم بترقية خزانة ملابسك مع هذا التصميم الأنيق من Hiccup وكن موضع حسد الجميع من حولك.'\n The attributes are:\n\t• Category: 'outwear > coatjacket'\n\t• Material: 'polyester'\n\t• Target Audience: women\n\t• Fabric Style: Corduroy\n\t• Pattern: Solid\n\t• Colar Type: Polo Collar\n\t• Sleeve Length: Long Sleeve\n\t• Details: Buttoned, Pocket\n\t• Tags: 'autumn, black, cosy, onsale, discount55_2403' (do not use any seasonal tags or other discount info nothing with '_' or color)\n \n\nI have a product called 'High Waist Skirt with a Belt'. The description reads: 'Step into style with our High Waist Skirt with a Belt in Brown. This chic and versatile skirt is designed to flatter your figure and elevate your fashion game. The high waistline accentuates your curves while the belt adds a touch of sophistication. Made from a premium blend of fabrics, this skirt is comfortable and durable, ensuring that it will become a staple in your wardrobe for years to come. Whether you're dressing up for a special occasion or keeping it casual for everyday wear, this High Waist Skirt with a Belt in Brown is the perfect choice. Embrace timeless elegance and make a statement with this must-have piece.'. Arabic naming: 'تنورة عالية الخصر مع حزام'. Arabic description: 'نقدم لك تنورة عالية الخصر مع حزام، وهي إضافة لا بد منها لأي خزانة ملابس مواكبة للموضة. ارتقي بأسلوبك مع هذه القطعة الأنيقة والمتعددة الاستخدامات التي تجمع بين الرقي والراحة بسهولة. يبرز تصميم الخصر العالي منحنياتك، بينما يشد الحزام المتضمن خصرك للحصول على مظهر جذاب. مصنوعة من قماش عالي الجودة، هذه التنورة تنسدل بشكل جميل ومثالية للمناسبات غير الرسمية والرسمية. ارتديه مع بلوزة وحذاء بكعب لإطلالة أنيقة، أو ارتديه مع تي شيرت وحذاء رياضي لإطلالة أكثر استرخاءً. تمتعي بالأناقة الخالدة مع تنورتنا ذات الخصر العالي مع الحزام واخرجي بأناقة.'\n The attributes are:\n\t• Category: 'bottom > skirt'\n\t• Material: 'polyester'\n\t• Target Audience: women\n\t• Fabric Style: Woven\n\t• Waist: High Waist\n\t• Style: Mini\n\t• Details: Belt\n\t• Tags: 'black, springsummer, allnewin, neutral, tailoring' (do not use any seasonal tags or other discount info nothing with '_' or color)\n \n\nI have a product called 'Knitted Crew Neck Crop Top'. The description reads: 'Introducing our Knitted Crew Neck Crop Top, the perfect addition to your wardrobe. Made from high-quality knitted fabric, this crop top offers both style and comfort. The crew neck design adds a touch of sophistication, while the cropped length adds a modern twist. Whether you pair it with high-waisted jeans for a casual look or dress it up with a skirt for a night out, this crop top is versatile and effortlessly chic. The soft fabric feels luxurious against the skin, making it a pleasure to wear all day long. Upgrade your style game with our Knitted Crew Neck Crop Top and turn heads wherever you go.'. Arabic naming: 'محبوك طاقم الرقبة المحاصيل الأعلى'. \nArabic description: 'نقدم لك القميص القصير ذو الرقبة الدائرية المحبوك، الإضافة المثالية لخزانة ملابسك. مصنوع من قماش محبوك عالي الجودة، هذا القميص القصير يوفر الأناقة والراحة. يضيف تصميم الرقبة الدائرية لمسة من الرقي، بينما يضيف الطول القصير لمسة عصرية. سواء كنت ترتديه مع بنطال جينز عالي الخصر لمظهر غير رسمي أو ترتديه مع تنورة لقضاء ليلة في الخارج، فإن هذا القميص القصير متعدد الاستخدامات وأنيق دون عناء. يعطي النسيج الناعم ملمسًا فاخرًا على البشرة، مما يجعل ارتداؤه ممتعًا طوال اليوم. قم بترقية أسلوبك في الأناقة مع القميص المحبوك ذو الرقبة الدائرية وجذب الأنظار أينما ذهبت.'\n The attributes are:\n\t• Category: 'top > croptop'\n\t• Material: 'viscose, polyamide'\n\t• Target Audience: women\n\t• Fabric Style: Knitted\n\t• Pattern: Solid\n\t• Colar Type: Crew Neck\n\t• Sleeve Length: Sleeveless\n\t• Tags: 'autumn, neutral' (do not use any seasonal tags or other discount info nothing with '_' or color)\n " def shot(input, category, level): output_dict = {} if level == 'variant': openai_parsed_response = get_openAI_tags(ast.literal_eval(str(input))) face_embeddings = get_face_embeddings(ast.literal_eval(str(input))) cropped_images, product_crops = get_cropped_images(ast.literal_eval(str(input)), category) print(product_crops) if len(product_crops) == 0: print("No product crops found. Using image urls instead.") subColour, mainColour, score = get_colour(ast.literal_eval(str(input)), category) else: subColour, mainColour, score = get_colour(product_crops, category) # Ensure all outputs are JSON serializable output_dict['colors'] = { "main": mainColour, "sub": subColour, "score": score } output_dict['image_mapping'] = openai_parsed_response output_dict['face_embeddings'] = face_embeddings output_dict['cropped_images'] = cropped_images if level == 'product': common_result = get_predicted_attributes(ast.literal_eval(str(input)), category) output_dict['attributes'] = common_result output_dict['subcategory'] = category print(common_result) output_dict['description'] = get_product_description(category=category, attributes=common_result) # # Convert the dictionary to a JSON-serializable format # try: # serialized_output = json.dumps(output_dict) # except TypeError as e: # print(f"Serialization Error: {e}") # return {"error": "Serialization failed"} return json.dumps(output_dict) # @spaces.GPU # def get_colour(image_urls, category): # colourLabels = list(COLOURS_DICT.keys()) # for i in range(len(colourLabels)): # colourLabels[i] = colourLabels[i] + " clothing: " + category # responses = pipe(image_urls, candidate_labels=colourLabels) # # Get the most common colour # mainColour = responses[0][0]['label'].split(" clothing:")[0] # if mainColour not in COLOURS_DICT: # return None, None, None # # Add category to the end of each label # labels = COLOURS_DICT[mainColour] # for i in range(len(labels)): # labels[i] = labels[i] + " clothing: " + category # # Run pipeline in one go # responses = pipe(image_urls, candidate_labels=labels) # subColour = responses[0][0]['label'].split(" clothing:")[0] # return subColour, mainColour, responses[0][0]['score'] # @spaces.GPU # def get_colour(image_urls, category): # # Prepare color labels # colourLabels = [f"{color} clothing: {category}" for color in COLOURS_DICT.keys()] # print("Colour Labels:", colourLabels) # Debug: Print colour labels # print("Image URLs:", image_urls) # Debug: Print image URLs # # Split labels into two batches # mid_index = len(colourLabels) // 2 # first_batch = colourLabels[:mid_index] # second_batch = colourLabels[mid_index:] # # Process the first batch # responses_first_batch = pipe(image_urls, candidate_labels=first_batch) # # Get the top 3 from the first batch # top3_first_batch = sorted(responses_first_batch[0], key=lambda x: x['score'], reverse=True)[:3] # # Process the second batch # responses_second_batch = pipe(image_urls, candidate_labels=second_batch) # # Get the top 3 from the second batch # top3_second_batch = sorted(responses_second_batch[0], key=lambda x: x['score'], reverse=True)[:3] # # Combine the top 3 from each batch # combined_top6 = top3_first_batch + top3_second_batch # # Get the final top 3 from the combined list # final_top3 = sorted(combined_top6, key=lambda x: x['score'], reverse=True)[:3] # mainColour = final_top3[0]['label'].split(" clothing:")[0] # if mainColour not in COLOURS_DICT: # return None, None, None # # Get sub-colors for the main color # labels = [f"{label} clothing: {category}" for label in COLOURS_DICT[mainColour]] # print("Labels for pipe:", labels) # Debug: Confirm labels are correct # responses = pipe(image_urls, candidate_labels=labels) # subColour = responses[0][0]['label'].split(" clothing:")[0] # return subColour, mainColour, responses[0][0]['score'] @spaces.GPU def get_colour(image_urls, category): colour_labels = [f"{colour}, clothing: {category}" for colour in COLOURS_DICT.keys()] responses = pipe(image_urls, candidate_labels=colour_labels) main_colour, main_score = get_most_common_label(responses) if main_colour not in COLOURS_DICT: return None, None, None score = [main_score] labels = COLOURS_DICT[main_colour] if main_colour == "multicolor": labels = [label for key, values in COLOURS_DICT.items() if key != main_colour for label in values] labels = [f"{label}, clothing: {category}" for label in labels] responses = pipe(image_urls, candidate_labels=labels) most_common, sub_score = get_most_common_label(responses) sub_colours = [most_common] score.append(sub_score) if main_colour == "multicolor": sub_key = next(key for key, values in COLOURS_DICT.items() if most_common in values) labels = [label for key, values in COLOURS_DICT.items() if key not in {main_colour, sub_key} for label in values] labels = [f"{label}, clothing: {category}" for label in labels] responses = pipe(image_urls, candidate_labels=labels) most_common, tertiary_score = get_most_common_label(responses) sub_colours.append(most_common) score.append(tertiary_score) return sub_colours, main_colour, score # Function for get_predicted_attributes and get_colour def get_most_common_label(responses): feature_scores = defaultdict(float) for response in responses: label, score = response[0]['label'].split(", clothing:")[0], response[0]['score'] feature_scores[label] += score max_label = max(feature_scores, key=feature_scores.get) return max_label, feature_scores[max_label] / len(responses) @spaces.GPU def get_predicted_attributes(image_urls, category): # Assuming ATTRIBUTES_DICT and pipe are defined outside this function attributes = list(ATTRIBUTES_DICT.get(category, {}).keys()) # Mapping of possible values per attribute common_result = [] for attribute in attributes: values = ATTRIBUTES_DICT.get(category, {}).get(attribute, []) if len(values) == 0: continue # Adjust labels for the pipeline attribute = attribute.replace("colartype", "collar").replace("sleevelength", "sleeve length").replace("fabricstyle", "fabric") values = [f"{attribute}: {value.strip()}, clothing: {category}" for value in values] # Get the predicted values for the attribute responses = pipe(image_urls, candidate_labels=values) most_common, score = get_most_common_label(responses) common_result.append(most_common) if attribute == "details": # Process additional details labels if the score is higher than 0.8 for _ in range(2): values = [value for value in values if value != f"{most_common}, clothing: {category}"] responses = pipe(image_urls, candidate_labels=values) most_common, score = get_most_common_label(responses) if score > DETAILS_THRESHOLD * len(image_urls): common_result.append(most_common) # Convert common_result into a dictionary final = {} details_count = 0 for result in common_result: result = result.replace("collar", "colartype").replace("sleeve length", "sleevelength").replace("fabric", "fabricstyle") key, value = result.split(": ") if key == "details": if details_count > 0: key += str(details_count) details_count += 1 final[key] = value.lower() return final def generate_prompt(category: Optional[str], tags: Optional[Set[str]], materials: Optional[List[Dict[str, int]]], attributes: Optional[List[Dict[str, str]]]) -> str: print(attributes) for attr in attributes: print(attr) print(attributes[attr]) formatted_attributes = [f"{attr}: {attributes[attr]}" for attr in attributes] if attributes else [] formatted_string = "\\n".join(formatted_attributes) if formatted_attributes else "No attributes provided." processed_category = category.replace("women-", "").replace("-", " > ") if category else None material_keys = ", ".join([str(material['key']) for material in materials]) if materials else None processed_tags = ", ".join(tags) if tags else None return ( f"I have a new product.\\nThe attributes are: Category: {processed_category}\\n" f"Material: {material_keys}\\n{formatted_string}\\n" f"Tags: {processed_tags} (do not use any seasonal tags or other discount info nothing with '_' or color)\\n" "Print the output as a json object no extra text, keys are name_en, name_ar, summary_en, summary_ar" ).strip() def send_request_to_openai(prompt: str) -> Optional[Dict[str, Any]]: messages = [ {"role": "system", "content": "You are a labelling assistant, you will help create product metadata for my online e-commerce platform. Your tasks will be to return product names and product descriptions. You will be given some product attributes and/or metadata guideline outputs and are expected to follow the sample output to generate new outputs.\n\nYou will be asked to generate product names and descriptions in both english and arabic."}, {"role": "user", "content": "Hello, I need help creating product metadata for new apparel items. I will provide examples, I want you to understand the current structure and setup. You are expected to match the current tone of voice and overall information."}, {"role": "assistant", "content": "Sure, I'd be happy to help! Please provide the attributes and any existing product names you'd like me to consider"}, {"role": "user", "content": OPENAI_USER_PROMPT}, {"role": "assistant", "content": "This was very helpful, I have taken note of the overall data inputs and expected product name and description outputs. Please provide the attributes for the newly listed product so I can generate the name and description for you."}, {"role": "user", "content": prompt} ] try: openai_response = client.chat.completions.create( model="gpt-3.5-turbo-16k", messages=messages, temperature=1.0, max_tokens=512, top_p=1.0, frequency_penalty=0.0, presence_penalty=0.0 ) json_string = openai_response.choices[0].message.content parsed_response = json.loads(json_string) print(f"English name: {parsed_response['name_en']}") print(f"Arabic name: {parsed_response['name_ar']}") print(f"English description: {parsed_response['summary_en']}") print(f"Arabic description: {parsed_response['summary_ar']}") return parsed_response except openai.error.OpenAIError as e: print(f"Failed to fetch details: {e}") return None def get_product_description(category: Optional[str] = None, tags: Optional[Set[str]] = None, attributes: Optional[List[Dict[str, str]]] = None, materials: Optional[List[Dict[str, int]]] = None) -> Optional[Dict[str, Any]]: prompt = generate_prompt(category, tags, materials, attributes) return send_request_to_openai(prompt) # Example usage: # response = get_product_description(api_key, category="women-shirts", tags={"summer", "casual"}, attributes=[{"key": "color", "value": "blue"}], materials=[{"key": "cotton", "value": 100}]) def get_openAI_tags(image_urls): # Create list containing JSONs of each image URL imageList = [] for image in image_urls: imageList.append({"type": "image_url", "image_url": {"url": image}}) try: openai_response = client.chat.completions.create( model="gpt-4o", messages=[ { "role": "system", "content": [ { "type": "text", "text": "You're a tagging assistant, you will help label and tag product pictures for my online e-commerce platform. Your tasks will be to return which angle the product images were taken from. You will have to choose from 'full-body', 'half-body', 'side', 'back', or 'zoomed' angles. You should label each of the images with one of these labels depending on which you think fits best (ideally, every label should be used at least once, but only if there are 5 or more images), and should respond with an unformatted dictionary where the key is a string representation of the url index of the url and the value is the assigned label." } ] }, { "role": "user", "content": imageList }, ], temperature=1, max_tokens=500, top_p=1, frequency_penalty=0, presence_penalty=0 ) response = json.loads(openai_response.choices[0].message.content) return response except Exception as e: print(f"OpenAI API Error: {e}") return {} @spaces.GPU def get_face_embeddings(image_urls): # Initialize a dictionary to store the face encodings or errors results = {} # Loop through each image URL for index, url in enumerate(image_urls): try: # Try to download the image from the URL response = requests.get(url) # Raise an exception if the response is not successful response.raise_for_status() # Load the image using face_recognition image = face_recognition.load_image_file(BytesIO(response.content)) # Get the face encodings for all faces in the image face_encodings = face_recognition.face_encodings(image) # If no faces are detected, store an empty list if not face_encodings: results[str(index)] = [] else: # Otherwise, store the first face encoding as a list results[str(index)] = face_encodings[0].tolist() except Exception as e: # If any error occurs during the download or processing, store the error message print(f"Error processing image: {str(e)}") return results # new ACCURACY_THRESHOLD = 0.86 def open_image_from_url(url): # Fetch the image from the URL response = requests.get(url, stream=True) response.raise_for_status() # Check if the request was successful # Open the image using PIL image = Image.open(BytesIO(response.content)) return image # Add the main data to the session state main = [['Product Id', 'Sku', 'Color', 'Images', 'Status', 'Category', 'Text']] # This is the order of the categories list. NO NOT CHANGE. Just for visualization purposes cats = ['shirt, blouse', 'top, t-shirt, sweatshirt', 'sweater', 'cardigan', 'jacket', 'vest', 'pants', 'shorts', 'skirt', 'coat', 'dress', 'jumpsuit', 'cape', 'glasses', 'hat', 'headband, head covering, hair accessory', 'tie', 'glove', 'watch', 'belt', 'leg warmer', 'tights, stockings', 'sock', 'shoe', 'bag, wallet', 'scarf', 'umbrella', 'hood', 'collar', 'lapel', 'epaulette', 'sleeve', 'pocket', 'neckline', 'buckle', 'zipper', 'applique', 'bead', 'bow', 'flower', 'fringe', 'ribbon', 'rivet', 'ruffle', 'sequin', 'tassel'] filter = ['dress', 'jumpsuit', 'cape', 'glasses', 'hat', 'headband, head covering, hair accessory', 'tie', 'glove', 'watch', 'belt', 'leg warmer', 'tights, stockings', 'sock', 'shoe', 'scarf', 'umbrella', 'hood', 'collar', 'lapel', 'epaulette', 'sleeve', 'pocket', 'neckline', 'buckle', 'zipper', 'applique', 'bead', 'bow', 'flower', 'fringe', 'ribbon', 'rivet', 'ruffle', 'sequin', 'tassel'] # 0 for full body, 1 for upper body, 2 for lower body, 3 for over body (jacket, coat, etc), 4 for accessories yolo_mapping = { 'shirt, blouse': 3, 'top, t-shirt, sweatshirt' : 1, 'sweater': 1, 'cardigan': 1, 'jacket': 3, 'vest': 1, 'pants': 2, 'shorts': 2, 'skirt': 2, 'coat': 3, 'dress': 0, 'jumpsuit': 0, 'bag, wallet': 4 } # First line full body, second line upper body, third line lower body, fourth line over body, fifth line accessories label_mapping = [ ['women-dress-mini', 'women-dress-dress', 'women-dress-maxi', 'women-dress-midi', 'women-playsuitsjumpsuits-playsuit', 'women-playsuitsjumpsuits-jumpsuit', 'women-coords-coords', 'women-swimwear-onepieces', 'women-swimwear-bikinisets'], ['women-sweatersknits-cardigan', 'women-top-waistcoat', 'women-top-blouse', 'women-sweatersknits-blouse', 'women-sweatersknits-sweater', 'women-top-top', 'women-loungewear-hoodie', 'women-top-camistanks', 'women-top-tshirt', 'women-top-croptop', 'women-loungewear-sweatshirt', 'women-top-body'], ['women-loungewear-joggers', 'women-bottom-trousers', 'women-bottom-leggings', 'women-bottom-jeans', 'women-bottom-shorts', 'women-bottom-skirt', 'women-loungewear-activewear', 'women-bottom-joggers'], ['women-top-shirt', 'women-outwear-coatjacket', 'women-outwear-blazer', 'women-outwear-coatjacket', 'women-outwear-kimonos'], ['women-accessories-bags'] ] MODEL_NAME = "valentinafeve/yolos-fashionpedia" feature_extractor = YolosImageProcessor.from_pretrained('hustvl/yolos-small') model = YolosForObjectDetection.from_pretrained(MODEL_NAME) def get_category_index(category): # Find index of label mapping for i, labels in enumerate(label_mapping): if category in labels: break return i def get_yolo_index(category): # Find index of yolo mapping return yolo_mapping[category] def fix_channels(t): """ Some images may have 4 channels (transparent images) or just 1 channel (black and white images), in order to let the images have only 3 channels. I am going to remove the fourth channel in transparent images and stack the single channel in back and white images. :param t: Tensor-like image :return: Tensor-like image with three channels """ if len(t.shape) == 2: return ToPILImage()(torch.stack([t for i in (0, 0, 0)])) if t.shape[0] == 4: return ToPILImage()(t[:3]) if t.shape[0] == 1: return ToPILImage()(torch.stack([t[0] for i in (0, 0, 0)])) return ToPILImage()(t) def idx_to_text(i): return cats[i] # Random colors used for visualization COLORS = [[0.000, 0.447, 0.741], [0.850, 0.325, 0.098], [0.929, 0.694, 0.125], [0.494, 0.184, 0.556], [0.466, 0.674, 0.188], [0.301, 0.745, 0.933]] # for output bounding box post-processing def box_cxcywh_to_xyxy(x): x_c, y_c, w, h = x.unbind(1) b = [(x_c - 0.5 * w), (y_c - 0.5 * h), (x_c + 0.5 * w), (y_c + 0.5 * h)] return torch.stack(b, dim=1) def rescale_bboxes(out_bbox, size): img_w, img_h = size b = box_cxcywh_to_xyxy(out_bbox) b = b * torch.tensor([img_w, img_h, img_w, img_h], dtype=torch.float32) return b def plot_results(pil_img, prob, boxes): plt.figure(figsize=(16,10)) plt.imshow(pil_img) ax = plt.gca() colors = COLORS * 100 i = 0 crops = [] crop_classes = [] for p, (xmin, ymin, xmax, ymax), c in zip(prob, boxes.tolist(), colors): cl = p.argmax() # Save each box as an image box_img = pil_img.crop((xmin, ymin, xmax, ymax)) crops.append(box_img) crop_classes.append(idx_to_text(cl)) ax.add_patch(plt.Rectangle((xmin, ymin), xmax - xmin, ymax - ymin, fill=False, color=c, linewidth=3)) ax.text(xmin, ymin, idx_to_text(cl), fontsize=10, bbox=dict(facecolor=c, alpha=0.8)) i += 1 # Remove white padding all around the image plt.axis('off') plt.subplots_adjust(left=0, right=1, top=1, bottom=0) output_img = plt.gcf() plt.close() return output_img, crops, crop_classes def visualize_predictions(image, outputs, threshold=0.8): # Keep only predictions with confidence >= threshold probas = outputs.logits.softmax(-1)[0, :, :-1] keep = probas.max(-1).values > threshold # Convert predicted boxes from [0; 1] to image scales bboxes_scaled = rescale_bboxes(outputs.pred_boxes[0, keep].cpu(), image.size) # Get filtered probabilities and boxes based on the filter list filter_set = set(filter) filtered_probas_boxes = [ (proba, box) for proba, box in zip(probas[keep], bboxes_scaled) if idx_to_text(proba.argmax()) not in filter_set ] # If there is a jumpsuit or dress detected, remove them if there are other clothes detected contains_jumpsuit_or_dress = any(idx_to_text(proba.argmax()) in ["jumpsuit", "dress"] for proba, _ in filtered_probas_boxes) if contains_jumpsuit_or_dress and len(filtered_probas_boxes) > 1: filtered_probas_boxes = [ (proba, box) for proba, box in filtered_probas_boxes if idx_to_text(proba.argmax()) not in ["jumpsuit", "dress"] ] # Remove duplicates: Only keep one box per class unique_classes = set() unique_filtered_probas_boxes = [] for proba, box in filtered_probas_boxes: class_text = idx_to_text(proba.argmax()) if class_text not in unique_classes: unique_classes.add(class_text) unique_filtered_probas_boxes.append((proba, box)) # If there are remaining filtered probabilities, plot results output_img = None crops = None crop_classes = None if unique_filtered_probas_boxes: final_probas, final_boxes = zip(*unique_filtered_probas_boxes) output_img, crops, crop_classes = plot_results(image, list(final_probas), torch.stack(final_boxes)) # Return the classes of the detected objects return [proba.argmax().item() for proba, _ in unique_filtered_probas_boxes], output_img, crops, crop_classes @spaces.GPU def get_objects(image, threshold=0.8): class_counts = {} image = fix_channels(ToTensor()(image)) image = image.resize((600, 800)) inputs = feature_extractor(images=image, return_tensors="pt") outputs = model(**inputs) detected_classes, output_img, crops, crop_classes = visualize_predictions(image, outputs, threshold=threshold) for cl in detected_classes: class_name = idx_to_text(cl) if class_name not in class_counts: class_counts[class_name] = 0 class_counts[class_name] += 1 if crop_classes is not None: crop_classes = [get_yolo_index(c) for c in crop_classes] return class_counts, output_img, crops, crop_classes def encode_images_to_base64(cropped_list): base64_images = [] for image in cropped_list: with io.BytesIO() as buffer: image.convert('RGB').save(buffer, format='JPEG') base64_image = base64.b64encode(buffer.getvalue()).decode('utf-8') base64_images.append(base64_image) return base64_images # def get_cropped_images(images,category): # cropped_list = [] # resultsPerCategory = {} # for num, image in enumerate(images): # image = open_image_from_url(image) # class_counts, output_img, cropped_images, cropped_classes = get_objects(image, 0.35) # if not class_counts: # continue # # Get the inverse category as any other mapping label except the current one corresponding category # inverse_category = [label for i, labels in enumerate(label_mapping) for label in labels if i != get_category_index(category) and i != 0] # # If category is a cardigan, we don't recommend category indices 1 and 3 # if category == 'women-sweatersknits-cardigan': # inverse_category = [label for i, labels in enumerate(label_mapping) for label in labels if i != get_category_index(category) and i != 1 and i != 3] # for i, image in enumerate(cropped_images): # cropped_category = cropped_classes[i] # print(cropped_category, cropped_classes[i], get_category_index(category)) # specific_category = label_mapping[cropped_category] # if cropped_category == get_category_index(category): # continue # cropped_list.append(image) # base64_images = encode_images_to_base64(cropped_list) # return base64_images def get_cropped_images(images, category): cropped_list = [] product_crops = [] for num, image in enumerate(images): try: image = open_image_from_url(image) class_counts, output_img, cropped_images, cropped_classes = get_objects(image, 0.37) if not class_counts: continue for i, image in enumerate(cropped_images): cropped_list.append(image) # If the detected class is the same as the category, add the image to the product crops if cropped_classes[i] == get_category_index(category): product_crops.append(image) except Exception as e: print(f"Error processing image {num}: {e}") return [] # Convert cropped images to base64 strings base64_images = encode_images_to_base64(cropped_list) product_base64_images = encode_images_to_base64(product_crops) return base64_images, product_base64_images # Define the Gradio interface with the updated components iface = gr.Interface( fn=shot, inputs=[ gr.Textbox(label="Image URLs (starting with http/https) comma seperated "), gr.Textbox(label="Category"), gr.Textbox(label="Level; accepted 'variant' or 'product'") ], outputs="text", examples=[ [['https://d2q1sfov6ca7my.cloudfront.net/eyJidWNrZXQiOiAiaGljY3VwLWltYWdlLWhvc3RpbmciLCAia2V5IjogIlc4MDAwMDAwMTM0LU9SL1c4MDAwMDAwMTM0LU9SLTEuanBnIiwgImVkaXRzIjogeyJyZXNpemUiOiB7IndpZHRoIjogODAwLCAiaGVpZ2h0IjogMTIwMC4wLCAiZml0IjogIm91dHNpZGUifX19', 'https://d2q1sfov6ca7my.cloudfront.net/eyJidWNrZXQiOiAiaGljY3VwLWltYWdlLWhvc3RpbmciLCAia2V5IjogIlc4MDAwMDAwMTM0LU9SL1c4MDAwMDAwMTM0LU9SLTIuanBnIiwgImVkaXRzIjogeyJyZXNpemUiOiB7IndpZHRoIjogODAwLCAiaGVpZ2h0IjogMTIwMC4wLCAiZml0IjogIm91dHNpZGUifX19', 'https://d2q1sfov6ca7my.cloudfront.net/eyJidWNrZXQiOiAiaGljY3VwLWltYWdlLWhvc3RpbmciLCAia2V5IjogIlc4MDAwMDAwMTM0LU9SL1c4MDAwMDAwMTM0LU9SLTMuanBnIiwgImVkaXRzIjogeyJyZXNpemUiOiB7IndpZHRoIjogODAwLCAiaGVpZ2h0IjogMTIwMC4wLCAiZml0IjogIm91dHNpZGUifX19'], "women-top-shirt","variant"]], description="Add an image URL (starting with http/https) or upload a picture, and provide a list of labels separated by commas.", title="Full product flow" ) # Launch the interface iface.launch()