from transformers import DetrFeatureExtractor, DetrForObjectDetection import requests import torch feature_extractor = DetrFeatureExtractor.from_pretrained("facebook/detr-resnet-50") model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50") # Core Pkgs import time from json import load import streamlit as st import cv2 from PIL import Image,ImageEnhance import numpy as np from io import BytesIO from transformers import pipeline st.set_page_config(page_title="Do Transform Images", initial_sidebar_state = "auto" ) st.title("Image Transformation & Detection App") st.text("Build with Streamlit and OpenCV") face_cascade = cv2.CascadeClassifier('frecog/haarcascade_frontalface_default.xml') eye_cascade = cv2.CascadeClassifier('frecog/haarcascade_eye.xml') smile_cascade = cv2.CascadeClassifier('frecog/haarcascade_smile.xml') #@st_cache #od(): #obj_detector = pipeline('object-detection') #return obj_detector def detect_faces(our_image): new_img = np.array(our_image.convert('RGB')) img = cv2.cvtColor(new_img,1) gray = cv2.cvtColor(new_img, cv2.COLOR_BGR2GRAY) # Detect faces faces = face_cascade.detectMultiScale(gray, 1.1, 4) # Draw rectangle around the faces for (x, y, w, h) in faces: cv2.rectangle(img, (x, y), (x+w, y+h), (255, 0, 0), 2) return img,faces def detect_eyes(our_image): new_img = np.array(our_image.convert('RGB')) img = cv2.cvtColor(new_img,1) gray = cv2.cvtColor(new_img, cv2.COLOR_BGR2GRAY) eyes = eye_cascade.detectMultiScale(gray, 1.3, 5) for (ex,ey,ew,eh) in eyes: cv2.rectangle(img,(ex,ey),(ex+ew,ey+eh),(0,255,0),2) return img def detect_smiles(our_image): new_img = np.array(our_image.convert('RGB')) img = cv2.cvtColor(new_img,1) gray = cv2.cvtColor(new_img, cv2.COLOR_BGR2GRAY) # Detect Smiles smiles = smile_cascade.detectMultiScale(gray, 1.1, 4) # Draw rectangle around the Smiles for (x, y, w, h) in smiles: cv2.rectangle(img, (x, y), (x+w, y+h), (255, 0, 0), 2) return img def cartonize_image(our_image): new_img = np.array(our_image.convert('RGB')) img = cv2.cvtColor(new_img,1) gray = cv2.cvtColor(new_img, cv2.COLOR_BGR2GRAY) # Edges gray = cv2.medianBlur(gray, 5) edges = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 9, 9) #Color color = cv2.bilateralFilter(img, 9, 300, 300) #Cartoon cartoon = cv2.bitwise_and(color, color, mask=edges) return cartoon def cannize_image(our_image): new_img = np.array(our_image.convert('RGB')) img = cv2.cvtColor(new_img,1) img = cv2.GaussianBlur(img, (11, 11), 0) canny = cv2.Canny(img, 100, 150) return canny def detect_objects(im): inputs = feature_extractor(images=im, return_tensors="pt") outputs = model(**inputs) # convert outputs (bounding boxes and class logits) to COCO API target_sizes = torch.tensor([im.size[::-1]]) results = feature_extractor.post_process(outputs, target_sizes=target_sizes)[0] boxes = [] f=None for score, label, box in zip(results["scores"], results["labels"], results["boxes"]): box = [round(i, 2) for i in box.tolist()] # let's only keep detections with score > 0.9 if score > 0.9: st.success( f"Detected {model.config.id2label[label.item()]} with confidence " f"{round(score.item(), 3)} at location {box}" ) boxes.append(box) new_img = np.array(im.convert('RGB')) img = cv2.cvtColor(new_img,1) for (x, y, w, h) in boxes: cv2.rectangle(img,(int(x),int(y)),(int(w), int(h)), (0, 0, 255)) return st.image(img)#st.image(box) @st.cache def load_image(img): im = Image.open(img) return im activities = ["Detection","About"] choice = st.sidebar.selectbox("Select Activty",activities) def change_photo_state(): st.session_state["photo"]="done" uploaded_photo = st.file_uploader("Upload Image",type=['jpg','png','jpeg'], on_change=change_photo_state) camera_photo = st.camera_input("Take a photo", on_change=change_photo_state) if "photo" not in st.session_state: st.session_state["photo"]="not done" if choice == 'Detection': st.subheader("Process your images ...") if st.session_state["photo"]=="done": if uploaded_photo: our_image= load_image(uploaded_photo) if camera_photo: our_image= load_image(camera_photo) if uploaded_photo==None and camera_photo==None: our_image=load_image("image.jpg") enhance_type = st.sidebar.radio("Enhance Type",["Original","Gray-Scale","Contrast","Brightness","Blurring"]) if enhance_type == 'Gray-Scale': new_img = np.array(our_image.convert('RGB')) img = cv2.cvtColor(new_img,1) gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY) # st.write(new_img) st.image(gray) elif enhance_type == 'Contrast': c_rate = st.sidebar.slider("Contrast",0.5,3.5) enhancer = ImageEnhance.Contrast(our_image) img_output = enhancer.enhance(c_rate) st.image(img_output) elif enhance_type == 'Brightness': c_rate = st.sidebar.slider("Brightness",0.5,3.5) enhancer = ImageEnhance.Brightness(our_image) img_output = enhancer.enhance(c_rate) st.image(img_output) elif enhance_type == 'Blurring': new_img = np.array(our_image.convert('RGB')) blur_rate = st.sidebar.slider("Brightness",0.5,3.5) img = cv2.cvtColor(new_img,1) blur_img = cv2.GaussianBlur(img,(11,11),blur_rate) st.image(blur_img) elif enhance_type == 'Original': st.image(our_image,width=300) else: st.image(our_image,width=300) # Face Detection task = ["Detect_any_objects", "Faces","Smiles","Eyes","Cannize","Cartonize"] feature_choice = st.sidebar.selectbox("Find Features",task) if st.button("Process"): if feature_choice == 'Faces': result_img,result_faces = detect_faces(our_image) st.image(result_img) st.success("Found {} faces".format(len(result_faces))) elif feature_choice == 'Smiles': result_img = detect_smiles(our_image) st.image(result_img) elif feature_choice == 'Eyes': with st.spinner('Wait for it...'): time.sleep(5) result_img = detect_eyes(our_image) st.image(result_img) elif feature_choice == 'Cartonize': result_img = cartonize_image(our_image) st.image(result_img) elif feature_choice == 'Cannize': result_canny = cannize_image(our_image) st.image(result_canny) elif feature_choice == 'Detect_any_objects': detect_objects(our_image) elif choice == 'About': st.subheader("About Face Detection App") st.markdown("Built with Streamlit by [Soumen Sarker](https://soumen-sarker-personal-website.streamlitapp.com/)") st.markdown("Credit [here](https://huggingface.co/models?pipeline_tag=object-detection)") #st.success("Isshor Saves @Soumen Sarker")