Dataset for YOLOv4

Dataset files and formats

• class_id is an integer greater than or equal to 0.
• center_x, center_y, widthand height are between 0.0 and 1.0.

converted-coco

image_path_and_bboxes.txt

<relative or full path>/image_0.jpg <class_id>,<center_x>,<center_y>,<width>,<height> <class_id>,<center_x>,<center_y>,<width>,<height> ...
<relative or full path>/image_1.jpg <class_id>,<center_x>,<center_y>,<width>,<height> ...
<relative or full path>/image_2.jpg <class_id>,<center_x>,<center_y>,<width>,<height> ...

...

<full path>
├── image_0.jpg
├── image_1.jpg
├── image_2.jpg
└── ...

Ref: https://github.com/hhk7734/tensorflow-yolov4/tree/master/test/dataset

yolo

image_path.txt

<relative or full path>/image_0.jpg
<relative or full path>/image_1.jpg
<relative or full path>/image_2.jpg

...

<image_name>.txt

<class_id> <center_x> <center_y> <width> <height>
<class_id> <center_x> <center_y> <width> <height>
<class_id> <center_x> <center_y> <width> <height>

...

<full path>
├── image_0.jpg
├── image_0.txt
├── image_1.jpg
├── image_1.txt
├── image_2.jpg
├── image_2.txt
└── ...

Convert coco to custom dataset

COCO 2017 Dataset

{
    "info": {
        "description": "COCO 2017 Dataset",
        "url": "http://cocodataset.org",
        "version": "1.0",
        "year": 2017,
        "contributor": "COCO Consortium",
        "date_created": "2017/09/01"
    },
    "licenses": [
        {
            "url": "http://creativecommons.org/licenses/by-nc-sa/2.0/",
            "id": 1,
            "name": "Attribution-NonCommercial-ShareAlike License"
        },
        ...
    ],
    "images": [
        {
            "license": 4,
            "file_name": "000000397133.jpg",
            "coco_url": "http://images.cocodataset.org/val2017/000000397133.jpg",
            "height": 427,
            "width": 640,
            "date_captured": "2013-11-14 17:02:52",
            "flickr_url": "http://farm7.staticflickr.com/6116/6255196340_da26cf2c9e_z.jpg",
            "id": 397133
        },
        ...
    ],
    "annotations": [
        {
            "segmentation": [[...]],
            "area": 702.1057499999998,
            "iscrowd": 0,
            "image_id": 289343,
            "bbox": [473.07,395.93,38.65,28.67], // xmin, ymin, width, height
            "category_id": 18,
            "id": 1768
        },
        ...
    ],
    "categories": [
        {
            "supercategory": "person",
            "id": 1,
            "name": "person"
        },
        {
            "supercategory": "vehicle",
            "id": 2,
            "name": "bicycle"
        },
        ...
    ]
}

Create names file

Create a file with names you want to predict. The class name should be included in categories above.

custom.names

person
bicycle
car
motorbike
aeroplane
bus

Conversion script

• 2020-10-16
• OS: Ubuntu 20.04

import json
from collections import OrderedDict
from tqdm import tqdm

INSTANCES_PATH = "instances_train2017.json"
NAMES_PATH = "custom.names"
OUTPUT_FILE_PATH = "custom_train2017.txt"

coco = json.load(open(INSTANCES_PATH))
images = coco["images"]
annotations = coco["annotations"]
categories = coco["categories"]
replaced_name = {
    "couch": "sofa",
    "airplane": "aeroplane",
    "tv": "tvmonitor",
    "motorcycle": "motorbike",
}

class_to_id = {}
id_to_class = {}
with open(NAMES_PATH, "r") as fd:
    index = 0
    for class_name in fd:
        class_name = class_name.strip()
        if len(class_name) != 0:
            id_to_class[index] = class_name
            class_to_id[class_name] = index
            index += 1

dataset = {}

for annotation in tqdm(annotations, desc="Parsing"):
    image_id = annotation["image_id"]
    category_id = annotation["category_id"]

    # Find image
    file_name = None
    image_height = 0
    image_width = 0
    for image in images:
        if image["id"] == image_id:
            file_name = image["file_name"]
            image_height = image["height"]
            image_width = image["width"]
            break

    if file_name is None:
        continue

    # Find class id
    class_id = None
    for category in categories:
        if category["id"] == category_id:
            category_name = category["name"]
            if category_name in replaced_name:
                category_name = replaced_name[category_name]

            class_id = class_to_id.get(category_name)
            break

    if class_id is None:
        continue

    # Calculate x,y,w,h
    x_center = annotation["bbox"][0] + annotation["bbox"][2] / 2
    x_center /= image_width
    y_center = annotation["bbox"][1] + annotation["bbox"][3] / 2
    y_center /= image_height
    width = annotation["bbox"][2] / image_width
    height = annotation["bbox"][3] / image_height

    if dataset.get(image_id):
        dataset[image_id][1].append(
            [class_id, x_center, y_center, width, height]
        )
    else:
        dataset[image_id] = [
            file_name,
            [[class_id, x_center, y_center, width, height]],
        ]

dataset = OrderedDict(sorted(dataset.items()))

with open(OUTPUT_FILE_PATH, "w") as fd:
    for image_id, bboxes in tqdm(dataset.items(), desc="Saving"):
        data = bboxes[0]
        for bbox in bboxes[1]:
            data += " "
            data += "{:d},".format(bbox[0])
            data += "{:8.6f},".format(bbox[1])
            data += "{:8.6f},".format(bbox[2])
            data += "{:8.6f},".format(bbox[3])
            data += "{:8.6f}".format(bbox[4])

        data += "\n"
        fd.write(data)

Dataset test script

v3

• 2021-02-21
• OS: Ubuntu 20.04
• yolov4: v3.1.0

import cv2
import numpy as np
from yolov4.tf import YOLOv4, YOLODataset

yolo = YOLOv4()
yolo.config.parse_names("test/coco.names")
yolo.config.parse_cfg("config/yolov4-tiny.cfg")

dataset = YOLODataset(
    config=yolo.config,
    dataset_list="/home/hhk7734/NN/val2017.txt",
    image_path_prefix="/home/hhk7734/NN/val2017",
)

count = 0
for i, (images, gt) in enumerate(dataset):
    cv2.namedWindow("truth", cv2.WINDOW_AUTOSIZE)

    beta_nms = yolo.config.metayolos[0].beta_nms
    classes = yolo.config.metayolos[0].classes
    stride = classes + 5
    num_mask = len(yolo.config.masks[0])

    candidates = []
    for y, metayolo in enumerate(yolo.config.metayolos):
        candidates.append(gt[y][..., : -len(yolo.config.yolo_0.mask)])

    for batch in range(len(images)):
        frame = images[batch, ...] * 255
        frame = cv2.cvtColor(frame.astype(np.uint8), cv2.COLOR_RGB2BGR)
        height, width, _ = frame.shape

        yolos = []
        for c, cand in enumerate(candidates):
            truth = cand[batch : batch + 1, ...]
            _, h, w, ch = truth.shape
            mask = yolo.config.masks[c]
            for n in range(num_mask):
                xy_index = n * stride
                wh_index = xy_index + 2
                oc_index = xy_index + 4

                xy = truth[..., xy_index:wh_index] * np.array([w, h])
                truth[..., xy_index:wh_index] = xy - xy.astype(np.int)

                awh = yolo.config.anchors[mask[n]]
                wh = truth[..., wh_index:oc_index] / awh
                truth[..., wh_index:oc_index] = np.log(
                    wh + np.finfo(np.float32).eps
                )

        treu_bboxes = yolo.get_yolo_detections(
            [c[batch : batch + 1, ...] for c in candidates]
        )
        # yolo.fit_to_original(treu_bboxes, height, width)
        image2 = yolo.draw_bboxes(frame, treu_bboxes)
        cv2.imshow("truth", image2)
        while cv2.waitKey(10) & 0xFF != ord("q"):
            pass
        count += 1
        if count == 30:
            break
    if count == 30:
        break

v2

• 2020-10-27
• OS: Ubuntu 20.04
• yolov4: v2.0.0

import cv2
import numpy as np
import tensorflow as tf
from yolov4.tf import YOLOv4

yolo = YOLOv4()
yolo.classes = "coco.names"
yolo.input_size = (640, 480)
yolo.batch_size = 2
dataset = yolo.load_dataset("train2017.txt", image_path_prefix="train2017")

for i, (images, gt) in enumerate(dataset):
    for j in range(len(images)):
        _candidates = []
        for candidate in gt:
            grid_size = candidate.shape[1:3]
            _candidates.append(
                tf.reshape(
                    candidate[j], shape=(1, grid_size[0] * grid_size[1] * 3, -1)
                )
            )
        candidates = np.concatenate(_candidates, axis=1)

        frame = images[j, ...] * 255
        frame = frame.astype(np.uint8)

        pred_bboxes = yolo.candidates_to_pred_bboxes(candidates[0])
        pred_bboxes = yolo.fit_pred_bboxes_to_original(pred_bboxes, frame.shape)
        frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
        image = yolo.draw_bboxes(frame, pred_bboxes)
        cv2.namedWindow("result", cv2.WINDOW_AUTOSIZE)
        cv2.imshow("result", image)
        while cv2.waitKey(10) & 0xFF != ord("q"):
            pass
    if i == 10:
        break

cv2.destroyWindow("result")