yolov3-tf 0.1.3

YOLOv3 implementation in TensorFlow 2.x

YOLOv3-TF

YOLOv3 implementation in TensorFlow 2.x

Installation

pip install yolov3-tf

Depends on tensorflow >=2.3.0 <=2.9.1

Usage

The package consists of three core modules -

• dataset
• models
• utils

Dataset

The dataset.py module is for loading and transforming the tfrecords for object detection. The examples in the input tfrecords must match the parsing schema.

import yolov3_tf.dataset as dataset
train_dataset = dataset.load_tfrecord_dataset(tfrecords_path)
train_dataset = train_dataset.batch(batch_size)
train_dataset = train_dataset.map(
    lambda x, y: (
        dataset.transform_images(x, image_dim),
        dataset.transform_targets(y, anchors, anchor_masks, image_dim),
    )
)

Models

The models.py module consists of implementation of two YOLOv3 and YOLOv3 tiny in Tesnsorflow.

from yolov3_tf.models import YoloV3, YoloV3Tiny
model = YoloV3(image_dim = 416, training=True, classes=10)

Utils

The utils.py module provides some common functions for training YOLOv3 model, viz., loading weights, freezing layers, drawing boxes on images, compute iou

# convert weights 
from yolov3_tf.models import YoloV3, YoloV3Tiny
from yolov3_tf import utils

yolo = YoloV3()
utils.load_darknet_weights(yolo, weights_path, is_tiny=False)
yolo.save_weights(converted_weights_path)