run_video_trt.py

import argparse
import cv2
import numpy as np
import os
import torch
import torch.nn.functional as F
from torchvision.transforms import Compose
import time

import sys
sys.path.append("Depth_Anythingv2_TensorRT_python/models")

from depth_anything.dpt import DepthAnything
from depth_anything.util.transform import Resize, NormalizeImage, PrepareForNet

from Depth_Anythingv2_TensorRT_python.models.dpt import Dpt

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--video_path', type=str)
    parser.add_argument('--outdir', type=str, default='./vis_video_depth')
    parser.add_argument(
        "--engine", type=str, required=True, help="Path to the TensorRT engine"
    )
    
    args = parser.parse_args()
    
    margin_width = 50

    DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
    
    # depth_anything = DepthAnything.from_pretrained('LiheYoung/depth_anything_{}14'.format(args.encoder)).to(DEVICE).eval()
    
    # total_params = sum(param.numel() for param in depth_anything.parameters())
    # print('Total parameters: {:.2f}M'.format(total_params / 1e6))
    
    model = Dpt(args)
    
    # transform = Compose([
    #     Resize(
    #         width=518,
    #         height=518,
    #         resize_target=False,
    #         keep_aspect_ratio=True,
    #         ensure_multiple_of=14,
    #         resize_method='lower_bound',
    #         image_interpolation_method=cv2.INTER_CUBIC,
    #     ),
    #     NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    #     PrepareForNet(),
    # ])

    if os.path.isfile(args.video_path):
        if args.video_path.endswith('txt'):
            with open(args.video_path, 'r') as f:
                lines = f.read().splitlines()
        else:
            filenames = [args.video_path]
    else:
        filenames = os.listdir(args.video_path)
        filenames = [os.path.join(args.video_path, filename) for filename in filenames if not filename.startswith('.')]
        filenames.sort()
    
    os.makedirs(args.outdir, exist_ok=True)
    
    for k, filename in enumerate(filenames):
        print('Progress {:}/{:},'.format(k+1, len(filenames)), 'Processing', filename)
        
        raw_video = cv2.VideoCapture(filename)
        frame_width, frame_height = int(raw_video.get(cv2.CAP_PROP_FRAME_WIDTH)), int(raw_video.get(cv2.CAP_PROP_FRAME_HEIGHT))
        frame_rate = int(raw_video.get(cv2.CAP_PROP_FPS))
        output_width = frame_width * 2 + margin_width
        
        filename = os.path.basename(filename)
        output_path = os.path.join(args.outdir, filename[:filename.rfind('.')] + '_video_depth.mp4')
        out = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*"mp4v"), frame_rate, (output_width, frame_height))

        frame_id = 0
        total_inf_time = 0
        while raw_video.isOpened():
            ret, raw_frame = raw_video.read()
            if not ret:
                break
            
            frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2RGB)

            # frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2RGB) / 255.0
            
            # frame = transform({'image': frame})['image']
            # frame = torch.from_numpy(frame).unsqueeze(0).to(DEVICE)

            pro_image, shape_info = model.preprocess(frame)
            
            if frame_id == 0:
                print(f'frame: {pro_image.shape}')   
            start = time.time()            
            with torch.no_grad():
                # depth = depth_anything(frame)
                depth = model.inference(pro_image)
            end = time.time()
            if frame_id == 0:
                print(f'depth: {depth.shape}')
            inf_time = end - start
            # print(f"Inf time: {inf_time:.5f} sec")
            total_inf_time += inf_time

            depth = model.postprocess(shape_info, depth)

            # depth = F.interpolate(depth[None], (frame_height, frame_width), mode='bilinear', align_corners=False)[0, 0]
            # depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0

            depth_int = depth.astype(np.uint8)
            depth_color = cv2.applyColorMap(depth_int, cv2.COLORMAP_INFERNO)
            
            split_region = np.ones((frame_height, margin_width, 3), dtype=np.uint8) * 255
            combined_frame = cv2.hconcat([raw_frame, split_region, depth_color])
            
            out.write(combined_frame)

            frame_id += 1

        print(f'Avg inf time: {total_inf_time / frame_id}')
        raw_video.release()
        out.release()