Spaces:
Sleeping
Sleeping
Download models/feature_extractor.py from kabeerahmad/Feature_Extraction: direct link, hf CLI and curl.
- Browser
- Download file 3.67 kB
-
https://huggingface.co/spaces/kabeerahmad/Feature_Extraction/resolve/main/models/feature_extractor.py
- Command line
-
hf download hf://spaces/kabeerahmad/Feature_Extraction/models/feature_extractor.py
-
curl -L -o feature_extractor.py https://huggingface.co/spaces/kabeerahmad/Feature_Extraction/resolve/main/models/feature_extractor.py
3.67 kB
| import os | |
| import torch | |
| import clip # Ensure you installed via: pip install git+https://github.com/openai/CLIP.git | |
| from PIL import Image | |
| import numpy as np | |
| import cv2 | |
| import mediapipe as mp | |
| import matplotlib.pyplot as plt | |
| # Set device and load the CLIP model (CLIP automatically downloads the correct weights) | |
| device = "cuda" if torch.cuda.is_available() else "cpu" | |
| model, preprocess = clip.load("ViT-B/32", device=device) | |
| def isolate_subject(pil_image: Image.Image) -> Image.Image: | |
| """ | |
| Isolates the subject from the background using Mediapipe. | |
| """ | |
| image_np = np.array(pil_image) | |
| image_bgr = cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR) | |
| mp_selfie_segmentation = mp.solutions.selfie_segmentation | |
| with mp_selfie_segmentation.SelfieSegmentation(model_selection=1) as segmenter: | |
| results = segmenter.process(cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)) | |
| mask = results.segmentation_mask | |
| condition = mask > 0.5 # Threshold segmentation | |
| bg_color = (255, 255, 255) | |
| output_image = np.where(condition[..., None], image_np, bg_color) | |
| output_image = output_image.astype(np.uint8) | |
| return Image.fromarray(output_image) | |
| def zero_shot_classification(image_features, candidate_texts): | |
| """ | |
| Classify image features against candidate texts and return the best matching candidate. | |
| """ | |
| text_tokens = clip.tokenize(candidate_texts).to(device) | |
| with torch.no_grad(): | |
| text_features = model.encode_text(text_tokens) | |
| text_features /= text_features.norm(dim=-1, keepdim=True) | |
| similarities = (image_features @ text_features.T).squeeze(0).cpu().numpy() | |
| best_idx = int(np.argmax(similarities)) | |
| return candidate_texts[best_idx] | |
| def extract_skin_tone(pil_image: Image.Image): | |
| """ | |
| Extracts the average RGB color from the central region of the image. | |
| """ | |
| image = np.array(pil_image) | |
| h, w, _ = image.shape | |
| x_start, y_start = w // 4, h // 4 | |
| x_end, y_end = x_start + w // 2, y_start + h // 2 | |
| face_region = image[y_start:y_end, x_start:x_end] | |
| avg_color = face_region.mean(axis=(0, 1)) | |
| return tuple(int(c) for c in avg_color) | |
| def extract_features(image_path: str) -> dict: | |
| """ | |
| Process the image and extract features including: | |
| - Gender, Body Type, Hair Type, Hair Color, | |
| - Eyeball Color, Glasses status, and Skin Tone. | |
| """ | |
| pil_image = Image.open(image_path).convert("RGB") | |
| isolated_image = isolate_subject(pil_image) | |
| image_input = preprocess(isolated_image).unsqueeze(0).to(device) | |
| with torch.no_grad(): | |
| image_features = model.encode_image(image_input) | |
| image_features /= image_features.norm(dim=-1, keepdim=True) | |
| gender = zero_shot_classification(image_features, ["man", "woman"]) | |
| body_type = zero_shot_classification(image_features, ["slim", "moderate", "heavyset"]) | |
| hair_type = zero_shot_classification(image_features, ["bald", "short", "normal", "long"]) | |
| glasses = zero_shot_classification(image_features, ["wearing eyeglasses", "not wearing eyeglasses"]) | |
| glasses = "Yes" if glasses == "wearing eyeglasses" else "No" | |
| eyeball_color = zero_shot_classification(image_features, ["blue", "brown", "green", "hazel"]) | |
| hair_color = zero_shot_classification(image_features, ["blonde", "brown", "black", "red"]) | |
| skin_tone = extract_skin_tone(isolated_image) | |
| skin_tone_str = f"{skin_tone[0]},{skin_tone[1]},{skin_tone[2]}" | |
| return { | |
| "gender": gender, | |
| "body_type": body_type, | |
| "hair_type": hair_type, | |
| "hair_color": hair_color, | |
| "eyeball_color": eyeball_color, | |
| "glasses": glasses, | |
| "skin_tone": skin_tone_str, | |
| } | |