dnnlab

DnnLab

Dnnlab is a small framework for deep learning models based on TensorFlow.

It provides custom training loops for:

• Generative Models (GAN, cGan, cycleGAN)
• Image Detection (custom YOLO)

Additonaly custom Keras Layer:

• Non-Local-Blocks (Self-Attention)
• Squeeze and Excitation Blocks (SEBlocks)
• YOLO-Decoding Layer

Input pipeline functionality:

• YOLO (Tfrecords to Datasets)
• YOLO data augmentation
• Generative Models (Tfrecords to Datasets)

TensorBoard output:

• YOLO coco metrics (Precision (mAP) & Recall)
• YOLO loss (loss_class, loss_conf, loss_xywh, total_loss)
• YOLO bounding boxes
• Generative Models (Loss & Images)

Requirements

TensorFlow 2.3.0

Installation

Run the following to install:

pip install dnnlab