how to detect video

[WANG-1173]

If I want to test local video or webcam video, how should I modify it?

[PommesPeter]

I have the same question!
I have try to add the part of testing local video or webcam video code , but I meet some problems which I can't solve it.

[PommesPeter]

The problem is 'RuntimeError: Expected 4-dimensional input for 4-dimensional weight [32, 3, 3, 3], but got 3-dimensional input of size [480, 640, 3] instead' when I try to transform the frame image to tensor, but it occurs the error above.

[WANG-1173]

I have the same question!
I have try to add the part of testing local video or webcam video code , but I meet some problems which I can't solve it.
I also tried to add the test code, but because the weight problem could not solve the problem of the test video, I finally replaced someone else's yolov3 GitHub

[PommesPeter]

I think the way to modify the code is wrong, I also struggling for it.

I also find other yolov3 on Github, but it doesn't show in good quality.

[PommesPeter]

Could you tell me what yolov3 you are using? Thanks a lot

[WANG-1173]

Could you tell me what yolov3 you are using? Thanks a lot

git clone https://github.com/ultralytics/yolov3.git

[PommesPeter]

I appreciate you can share this one.
And I have a question, Is it you just replaced the file about video detect?

[WANG-1173]

I appreciate you can share this one.
And I have a question, Is it you just replaced the file about video detect?

I think you may have misunderstood what I mean, because I couldn't test the video, so I directly changed yolov3 and used the code of the link blogger above, which contains the command to test the video directly.

[PommesPeter]

alright.
Anyway, Thank you very much

[Guardian-Li]

from future import division

from models import *
from utils.utils import *
from utils.datasets import *

import os
import sys
import time
import datetime
import argparse
import cv2

from PIL import Image

import torch
from torch.utils.data import DataLoader
from torchvision import datasets
from torch.autograd import Variable

import matplotlib.pyplot as plt
import matplotlib.patches as patches
from matplotlib.ticker import NullLocator

if name == "main":
parser = argparse.ArgumentParser()
parser.add_argument("--image_folder", type=str, default="data/samples", help="path to dataset")
parser.add_argument("--vedio_file", type=str, default="vedio_samples/2.mp4", help="path to dataset")
parser.add_argument("--model_def", type=str, default="config/yolov3-tiny.cfg", help="path to model definition file")
parser.add_argument("--weights_path", type=str, default="model_trained/100-epoch-air.pth", help="path to weights file")
parser.add_argument("--class_path", type=str, default="data/air.names", help="path to class label file")
parser.add_argument("--conf_thres", type=float, default=0.8, help="object confidence threshold")
parser.add_argument("--nms_thres", type=float, default=0.4, help="iou thresshold for non-maximum suppression")
parser.add_argument("--batch_size", type=int, default=1, help="size of the batches")
parser.add_argument("--n_cpu", type=int, default=3, help="number of cpu threads to use during batch generation")
parser.add_argument("--img_size", type=int, default=416, help="size of each image dimension")
parser.add_argument("--checkpoint_model", type=str, help="path to checkpoint model")
opt = parser.parse_args()
print(opt)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = Darknet(opt.model_def, img_size=opt.img_size).to(device)
if opt.weights_path.endswith(".weights"):
# Load darknet weights
model.load_darknet_weights(opt.weights_path)
else:
# Load checkpoint weights
model.load_state_dict(torch.load(opt.weights_path))
model.cuda()
model.eval() # Set in evaluation mode
classes = load_classes(opt.class_path)
Tensor = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
if opt.vedio_file.endswith(".mp4"):
cap = cv2.VideoCapture(opt.vedio_file)
colors = np.random.randint(0, 255, size=(len(classes), 3), dtype="uint8")
a=[]
while cap.isOpened():
ret, img = cap.read()
PILimg = np.array(Image.fromarray(cv2.cvtColor(img,cv2.COLOR_BGR2RGB)))
imgTensor = transforms.ToTensor()(PILimg)
imgTensor, _ = pad_to_square(imgTensor, 0)
imgTensor = resize(imgTensor, 416)
imgTensor = imgTensor.unsqueeze(0)
imgTensor = Variable(imgTensor.type(Tensor))

    with torch.no_grad():
        detections = model(imgTensor)
        detections = non_max_suppression(detections, opt.conf_thres, opt.nms_thres)

    a.clear()
    if detections is not None:
        a.extend(detections)
    b=len(a)
    if len(a)  :
        for detections in a:
            if detections is not None:
                detections = rescale_boxes(detections, opt.img_size, PILimg.shape[:2])
                unique_labels = detections[:, -1].cpu().unique()
                n_cls_preds = len(unique_labels)
                for x1, y1, x2, y2, conf, cls_conf, cls_pred in detections:
                    box_w = x2 - x1
                    box_h = y2 - y1
                    color = [int(c) for c in colors[int(cls_pred)]]
                    print(cls_conf)
                    img = cv2.rectangle(img, (x1, y1 + box_h), (x2, y1), color, 2)
                    cv2.putText(img, classes[int(cls_pred)], (x1, y1), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
                    cv2.putText(img, str("%.2f" % float(conf)), (x2, y2 - box_h), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
                                color, 2)

        print()
        print()
    #cv2.putText(img,"Hello World!",(400,50),cv2.FONT_HERSHEY_PLAIN,2.0,(0,0,255),2)
    cv2.imshow('frame', img)
    #cv2.waitKey(0)

    if cv2.waitKey(25) & 0xFF == ord('q'):
        break
cap.release()
cv2.destroyAllWindows()

I write a simple vedio base on this

[PommesPeter]

@Guardian-Li It looks like to be similar to my code, but how do you solve the problem called:
RuntimeError: Expected 4-dimensional input for 4-dimensional weight [32, 3, 3, 3], but got 3-dimensional input of size [480, 640, 3] instead?

[PommesPeter]

@Guardian-Li almost the same code
the difference is that I haven't resized the frame from the camera

[PommesPeter]

@Guardian-Li Is the frame must be resized?

[Guardian-Li]

the Tensor you change from image must be use imgTensor = imgTensor.unsqueeze(0)

[Guardian-Li]

you should add one dimension

[Guardian-Li]

and cv2.imread is BRG ,PIL image read is RGB .

[Guardian-Li]

你那个张量最后加到model里面的之前给他加一个维度就行了

[PommesPeter]

@Guardian-Li 原来是中国人😂 好的 谢谢你

[Guardian-Li]

我看你简介的 没事 能跑就行

[PommesPeter]

嗯 我去改改我的代码

[PommesPeter]

@Guardian-Li 可以了 可以使用视频检测了 感谢帮助！

[Guardian-Li]

没事

[impravin22]

@Guardian-Li @PommesPeter Did you solve it? I need the detect.py file for this repo to detect video.

[aditjha]

@Guardian-Li @PommesPeter is there a working detect.py that works for video? The python code posted above by @Guardian-Li, is that working code?

[PommesPeter]

@aditjha it works, I follow his/her code and my understanding of the code, and i make it.

[PommesPeter]

@impravin22 Yes, I solve this problem. I follow his/her(@Guardian-Li ) code and I make it

[PommesPeter]

@impravin22 it's a working code

[impravin22]

@PommesPeter Great. Can you please send your detect.py if you dont mind.

Thanks.

[aditjha]

@PommesPeter thank you for replying! Also, I am new to all this...using yolo for a project...so this detect.py allows for a recorded video to be inferenced correct?

[impravin22]

@PommesPeter Yeah I found it. It is as video.py in @Guardian-Li repo, isn't it?

Thank you very much

[PommesPeter]

@impravin22 yes, my code is the same as his/her code

[impravin22]

@PommesPeter Thank you very much

[PommesPeter]

@aditjha I also the new to this project, you can read the source code over and over again to understand how it works.

[PommesPeter]

@impravin22 My pleasure!

[aditjha]

@PommesPeter yes, I will do so! quick question, in @Guardian-Li repo, i can see a vedio.py, but is the detect.py specific for videos or is the detect.py the same as the main repo

[PommesPeter]

the detect.py have a different from video.py. In @Guardian-Li 's repo, his/her detect.py is the same as the main repo. the video.py is the key to dectect video or camera

[aditjha]

@PommesPeter thank you! if I am trying to implement this repo along with @Guardian-Li 's video.py, will simply downloading her video.py work with this repo? Or did you have to download her repo and use everything from her repo?

[PommesPeter]

@aditjha it can be used in every yolo detection.

[RisithPerera]

How @Guardian-Li code change to only detect humans?