app.py

import gradio as gr
import torch
import requests
from torchvision import transforms

from sampling_util import furthest_neighbours
from video_reader import video_reader

model = torch.load("model").eval()
avg_pool = nn.AdaptiveAvgPool2d((1, 1))


def predict(input_file):
	base_directory = os.getcwd()
    selected_directory = os.path.join(base_directory, "selected_images")
    if os.path.isdir(selected_directory):
        shutil.rmtree(selected_directory)
    os.mkdir(selected_directory)
    
    zip_path = os.path.join(input_file.split('/')[-1][:-4] + ".zip")
    
    mean = [0.3156024, 0.33569682, 0.34337464]
    std = [0.16568947, 0.17827448, 0.18925823]
    
    img_vecs = []
    with torch.no_grad():
		for fp_i, file_path in enumerate([input_file]):
		    for i, in_img in enumerate(video_reader(file_path,
		                                         targetFPS=9,
		                                         targetWidth=100,
		                                         to_rgb=True)):
		        in_img = (in_img.astype(np.float32) / 255.)                                 
		        in_img = (in_img - mean) / std
		        in_img = np.transpose(in_img, (0, 3, 1, 2))
		        in_img = torch.from_numpy(in_img)
		        encoded = avg_pool(model(in_img))[0, :, 0, 0].cpu().numpy()
		        img_vecs += [encoded]

	img_vecs = np.asarray(img_vecs)
	rv_indices, _ = furthest_neighbours(
        img_vecs,
        downsample_size,
        seed=0)
    indices = np.zeros((img_vecs.shape[0],))
    indices[np.asarray(rv_indices)] = 1
    
    global_ctr = 0
    for fp_i, file_path in enumerate([input_file]):
        for i, img in enumerate(video_reader(file_path,
                                             targetFPS=9,
                                             targetWidth=None,
                                             to_rgb=False)):
            if indices[global_ctr] == 1:
                cv2.imwrite(join(selected_directory, str(global_ctr) + ".jpg"), img)
            global_ctr += 1
            
    all_selected_imgs_path = [join(selected_directory, f) for f in listdir(selected_directory) if isfile(join(selected_directory, f))]
    if 0 < len(all_file_paths):
    zipf = zipfile.ZipFile(zip_path, 'w', zipfile.ZIP_DEFLATED)
    for i, f in enumerate(all_selected_imgs_path):
        zipf.write(f, basename(f))
    zipf.close()
	
	return zip_path

demo = gr.Interface(
	fn=predict, 
	inputs=gr.inputs.Video(label="Upload Video File"),
	outputs=gr.outputs.File(label="Zip"))
             
demo.launch()