how to use it in yolov5?

[WANGCHAO1996]

Is the result after NMS in YoloV5 passed to WBF for secondary screening?@ZFTurbo

[NOBUGW]

你好，请问你解决了吗？

[ZFTurbo]

I used it in Yolov5. What you need to do is to read all txt files with predictions created by Yolo and gather it in single CSV file or just numpy table. Then you can use WBF.

Here is my code:

# coding: utf-8
__author__ = 'ZFTurbo: https://kaggle.com/zfturbo'

import glob
import os

def convert_yolov5_preds(valid_files_dir, labels_dir, out_file):
    valid_files = glob.glob(valid_files_dir + '*.jpg')
    print('Total image files: {}'.format(len(valid_files)))
    files = glob.glob(labels_dir + '*.txt')
    print('Total labels files: {}'.format(len(files)))
    valid_ids = [os.path.basename(f)[:-4] for f in valid_files]
    out = open(out_file, 'w')
    out.write('image_id,label,conf,x1,x2,y1,y2\n')
    fixes = 0
    for f in files:
        image_id = os.path.basename(f)[:-4]
        in1 = open(f, 'r')
        lines = in1.readlines()
        in1.close()
        valid_ids.remove(image_id)
        for line in lines:
            arr = line.strip().split(' ')
            class_id = arr[0]
            x = float(arr[1])
            y = float(arr[2])
            w = float(arr[3])
            h = float(arr[4])
            x1 = x - (w / 2)
            x2 = x + (w / 2)
            y1 = y - (h / 2)
            y2 = y + (h / 2)
            if x1 < 0:
                fixes += 1
                x1 = 0
            if x2 > 1:
                fixes += 1
                x2 = 1
            if y1 < 0:
                fixes += 1
                y1 = 0
            if y2 > 1:
                fixes += 1
                y2 = 1
            conf = arr[5]
            pred_str = '{},{},{},{:.6f},{:.6f},{:.6f},{:.6f}\n'.format(image_id, str(class_id), conf, x1, x2, y1, y2)
            out.write(pred_str)

    print(len(valid_ids))

    # Output empty IDs
    for image_id in list(valid_ids):
        out.write('{},,,,,,\n'.format(image_id))

    out.close()
    print('Fixes: {}'.format(fixes))
    print('Result was written in: {}'.format(out_file))


if __name__ == '__main__':
    # Location of images
    valid_files_dir = './valid_imgs/'
    # Location of yolo v5 predictions
    labels_dir = 'yolov5x/valid_iou_0.45_02.1/labels/'
    # CSF-file to store results
    out_file = 'yolov5x_full_valid_iou_0.45_0.1.csv'
    convert_yolov5_preds(valid_files_dir, labels_dir, out_file)

[haoxue1215]

I used it in Yolov5. What you need to do is to read all txt files with predictions created by Yolo and gather it in single CSV file or just numpy table. Then you can use WBF.

Here is my code:

# coding: utf-8
__author__ = 'ZFTurbo: https://kaggle.com/zfturbo'

import glob
import os

def convert_yolov5_preds(valid_files_dir, labels_dir, out_file):
    valid_files = glob.glob(valid_files_dir + '*.jpg')
    print('Total image files: {}'.format(len(valid_files)))
    files = glob.glob(labels_dir + '*.txt')
    print('Total labels files: {}'.format(len(files)))
    valid_ids = [os.path.basename(f)[:-4] for f in valid_files]
    out = open(out_file, 'w')
    out.write('image_id,label,conf,x1,x2,y1,y2\n')
    fixes = 0
    for f in files:
        image_id = os.path.basename(f)[:-4]
        in1 = open(f, 'r')
        lines = in1.readlines()
        in1.close()
        valid_ids.remove(image_id)
        for line in lines:
            arr = line.strip().split(' ')
            class_id = arr[0]
            x = float(arr[1])
            y = float(arr[2])
            w = float(arr[3])
            h = float(arr[4])
            x1 = x - (w / 2)
            x2 = x + (w / 2)
            y1 = y - (h / 2)
            y2 = y + (h / 2)
            if x1 < 0:
                fixes += 1
                x1 = 0
            if x2 > 1:
                fixes += 1
                x2 = 1
            if y1 < 0:
                fixes += 1
                y1 = 0
            if y2 > 1:
                fixes += 1
                y2 = 1
            conf = arr[5]
            pred_str = '{},{},{},{:.6f},{:.6f},{:.6f},{:.6f}\n'.format(image_id, str(class_id), conf, x1, x2, y1, y2)
            out.write(pred_str)

    print(len(valid_ids))

    # Output empty IDs
    for image_id in list(valid_ids):
        out.write('{},,,,,,\n'.format(image_id))

    out.close()
    print('Fixes: {}'.format(fixes))
    print('Result was written in: {}'.format(out_file))


if __name__ == '__main__':
    # Location of images
    valid_files_dir = './valid_imgs/'
    # Location of yolo v5 predictions
    labels_dir = 'yolov5x/valid_iou_0.45_02.1/labels/'
    # CSF-file to store results
    out_file = 'yolov5x_full_valid_iou_0.45_0.1.csv'
    convert_yolov5_preds(valid_files_dir, labels_dir, out_file)

then？

[czczmr]

你好，请问你解决了吗？

之后要怎么做呢，从保存好的out_file 中读取数据使用你们的wbf；
请问有从两个csv文件里读取框坐标后使用wbf的demo吗

[czczmr]

之后要怎么做呢，从保存好的out_file 中读取数据使用你们的wbf；
请问有从两个csv文件里读取框坐标后使用wbf的demo吗

[zyfn]

Please, I have used WBF in Yolov5.Dataset is coco128.Device is GPU-1050ti .But the post processing time of every image is closely 1.5s. The time is normally?

[Pu-Zhengpeng]

Hello friends,csv file i have made,then how to use wbf ?thanks for your reply

[Pu-Zhengpeng]

之后要怎么做呢，从保存好的out_file 中读取数据使用你们的wbf； 请问有从两个csv文件里读取框坐标后使用wbf的demo吗

哥们你现在解决了吗?可以请教一下？