app.py

import matplotlib.pyplot as plt
import numpy as np
from six import BytesIO
from PIL import Image
import tensorflow as tf
from object_detection.utils import label_map_util
from object_detection.utils import visualization_utils as viz_utils
from object_detection.utils import ops as utils_op
import tarfile
import wget 
import gradio as gr
from huggingface_hub import snapshot_download
import os 
import cv2
from tqdm import tqdm

PATH_TO_LABELS = 'data/label_map.pbtxt'   
category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True)

def pil_image_as_numpy_array(pilimg):

    img_array = tf.keras.utils.img_to_array(pilimg)
    img_array = np.expand_dims(img_array, axis=0)
    return img_array
    
def load_image_into_numpy_array(path):
                                    
    image = None
    image_data = tf.io.gfile.GFile(path, 'rb').read()
    image = Image.open(BytesIO(image_data))
    return pil_image_as_numpy_array(image)            

def load_model(model_repo_id):
    download_dir = snapshot_download(model_repo_id)
    saved_model_dir = os.path.join(download_dir, "saved_model")
    detection_model = tf.saved_model.load(saved_model_dir)
    return detection_model


def predict(pilimg,Threshold):
    
    image_np = pil_image_as_numpy_array(pilimg)
    
    if Threshold is None or Threshold == 0:
        Threshold=threshold_d
    else:
        Threshold= float(Threshold)
        
    return predict2(image_np,Threshold),predict3(image_np,Threshold),Threshold

def predict2(image_np,Threshold):

    results = detection_model(image_np)
    
    # if Threshold is None or Threshold == 0:
    #     Threshold=threshold_d
        
    # different object detection models have additional results
    result = {key:value.numpy() for key,value in results.items()}
    
    label_id_offset = 0
    image_np_with_detections = image_np.copy()

    viz_utils.visualize_boxes_and_labels_on_image_array(
        image_np_with_detections[0],
        result['detection_boxes'][0],
        (result['detection_classes'][0] + label_id_offset).astype(int),
        result['detection_scores'][0],
        category_index,
        use_normalized_coordinates=True,
        max_boxes_to_draw=20,
        min_score_thresh=Threshold,#0.38,
        agnostic_mode=False,
        line_thickness=2)

    result_pil_img2 = tf.keras.utils.array_to_img(image_np_with_detections[0])
    
    return result_pil_img2


def predict3(image_np,Threshold):
    
    results = detection_model2(image_np)

    # different object detection models have additional results
    result = {key:value.numpy() for key,value in results.items()}
    
    label_id_offset = 0
    image_np_with_detections = image_np.copy()

    viz_utils.visualize_boxes_and_labels_on_image_array(
        image_np_with_detections[0],
        result['detection_boxes'][0],
        (result['detection_classes'][0] + label_id_offset).astype(int),
        result['detection_scores'][0],
        category_index,
        use_normalized_coordinates=True,
        max_boxes_to_draw=20,
        min_score_thresh=Threshold,#.38,
        agnostic_mode=False,
        line_thickness=2)

    result_pil_img4 = tf.keras.utils.array_to_img(image_np_with_detections[0])
    
    return result_pil_img4

# def detect_video(video):
#     # Create a video capture object
#     cap = cv2.VideoCapture(video)

#     # Process frames in a loop
#     while cap.isOpened():
#         ret, frame = cap.read()
#         if not ret:
#             break

#         # Expand dimensions since model expects images to have shape: [1, None, None, 3]
#         image_np_expanded = np.expand_dims(frame, axis=0)

#         # Run inference
#         output_dict = model(image_np_expanded)

#         # Extract detections
#         boxes = output_dict['detection_boxes'][0].numpy()
#         scores = output_dict['detection_scores'][0].numpy()
#         classes = output_dict['detection_classes'][0].numpy().astype(np.int64)

#         # Draw bounding boxes and labels
#         image_np_with_detections = viz_utils.visualize_boxes_and_labels_on_image_array(
#             frame,
#             boxes,
#             classes,
#             scores,
#             category_index,
#             use_normalized_coordinates=True,
#             max_boxes_to_draw=20,
#             min_score_thresh=.5,
#             agnostic_mode=False)

#         # Yield the processed frame
#         yield image_np_with_detections

#     # Release resources
#     cap.release()

a = os.path.join(os.path.dirname(__file__), "data/c_base_detected.mp4")  # Video
b = os.path.join(os.path.dirname(__file__), "data/c_tuned_detected.mp4")  # Video

# def video_demo(video1, video2):
#     return [video1, video2]

label_id_offset = 0    
threshold_d= 0.38
REPO_ID = "apailang/mytfodmodel"
detection_model = load_model(REPO_ID)
REPO_ID2 = "apailang/mytfodmodeltuned"
detection_model2 = load_model(REPO_ID2)

samples_folder = 'data'
# pil_image = Image.open(image_path)
# image_arr = pil_image_as_numpy_array(pil_image)

# predicted_img = predict(image_arr)
# predicted_img.save('predicted.jpg')

test1 = os.path.join(os.path.dirname(__file__), "data/test1.jpeg")  
test2 = os.path.join(os.path.dirname(__file__), "data/test2.jpeg")  
test3 = os.path.join(os.path.dirname(__file__), "data/test3.jpeg")  
test4 = os.path.join(os.path.dirname(__file__), "data/test4.jpeg")
test5 = os.path.join(os.path.dirname(__file__), "data/test5.jpeg")
test6 = os.path.join(os.path.dirname(__file__), "data/test6.jpeg")
test7 = os.path.join(os.path.dirname(__file__), "data/test7.jpeg")
test8 = os.path.join(os.path.dirname(__file__), "data/test8.jpeg")
test9 = os.path.join(os.path.dirname(__file__), "data/test9.jpeg")
test10 = os.path.join(os.path.dirname(__file__), "data/test10.jpeg")
test11 = os.path.join(os.path.dirname(__file__), "data/test11.jpeg")
test12 = os.path.join(os.path.dirname(__file__), "data/test12.jpeg")

base_image = gr.Interface(
    fn=predict,
#    inputs=[gr.Image(type="pil"),gr.Slider(minimum=0.01, maximum=1, value=0.38 ,label="Threshold",info="[not in used]to set prediction confidence threshold")],
    inputs=[gr.Image(type="pil"),gr.Slider(minimum=0.05, maximum=1,step=0.05,value=threshold_d ,label="To change default 0.38 prediction confidence Threshold. Range 0.05 to 1",info="Select any image below to start, you may amend threshold after first inference")],

    outputs=[gr.Image(type="pil",label="Base Model Inference"),gr.Image(type="pil",label="Tuned Model Inference"),gr.Textbox(label="Both images inferenced threshold")],
    title="Luffy and Chopper Head detection. SSD mobile net V2 320x320 trained with animated characters only",
    description="Upload a Image for prediction or click on below examples. Prediction confident is defaut to >38%, you may adjust after first inference",
    examples=
        [[test1],[test2],[test3],[test4],[test5],[test6],[test7],[test8],[test9],[test10],[test11],[test12],],
    cache_examples=True,examples_per_page=12 #,label="select image with 0.38 threshold to inference, you may amend threshold after inference"
    )

# tuned_image = gr.Interface(
#     fn=predict3,
#     inputs=gr.Image(type="pil"),
#     outputs=gr.Image(type="pil"),
#     title="Luffy and Chopper face detection on images. Result comparison of base vs tuned SSD mobile net V2 320x320",
#     description="Upload a Image for prediction or click on below examples. Mobile net tuned with data Augmentation. Prediction confident >38%",
#     examples=[[test1],[test2],[test3],[test4],[test5],[test6],[test7],[test8],[test9],[test10],[test11],[test12],],
#     cache_examples=True
#     )#.launch(share=True)

# a = os.path.join(os.path.dirname(__file__), "data/a.mp4")  # Video
# b = os.path.join(os.path.dirname(__file__), "data/b.mp4")  # Video
# c = os.path.join(os.path.dirname(__file__), "data/c.mp4")  # Video

# video_out_file = os.path.join(samples_folder,'detected' + '.mp4')

# stt_demo = gr.Interface(
#     fn=display_two_videos,
#     inputs=gr.Video(),
#     outputs=gr.Video(type="mp4",label="Detected Video"),
#     examples=[
#         [a],
#         [b],
#         [c],
#     ],
#     cache_examples=False
# )


video = gr.Interface(
    fn=lambda x,y: [x,y], #video_demo,
    inputs=[gr.Video(label="Base Model Video",interactive=False),gr.Video(label="Tuned Model Video",interactive=False)], 
    outputs=[gr.Video(label="Base Model Inferenced Video"), gr.Video(label="Tuned Model Inferenced Video")],
    examples=[
        [a, b]
    ],
    title="Luffy and Chopper face detection on video Result comparison of base vs tuned SSD mobile net V2 320x320",
    description="Model has been customed trained to detect Character of Luffy and Chopper with Prediction confident >10%. Videos are pre-inferenced to reduce load time. (Browser zoom out to view right columne - top (base model inference) & bottom(tuned model inference)) "
)

demo = gr.TabbedInterface([base_image, video], ["Images", "Video"])


if __name__ == "__main__":
    demo.launch()