YOLOv4_DarkNet

• This is a project that I am working on, at NCRAI in collaboration with Kerala Agricultural University. I will be working on the machine learning part of the project. I have created an object detection model using Yolo V4 algorithm with Darknet framework .

• You can download the YOLO V4 paper here

Index of this Repository

• The yolov4_custom_object_detection.ipynb is the code to train yolov4 using darknet

• The folder yolov4_test contains the following data :

1. obg.zip (annotated training image data)

2. test.zip (annotated testing image data)

3. generate_train.py (python code to generate train.txt file)

4. generate_test.py (python code to generate test.txt file)

5. obj.names (names of the classe(s) )

6. obj.data (contains informations like class, location of train.txt&test.txt)

7. yolov4-obj.cfg (contains various arguments like convolutional layers, filters, max_batch,various augmentation parameters...)

Problem

Detect and classify two classes of Pineapples

1. MATURED PINEAPPLE

2. UNMATURED PINEAPPLE

Dataset

The Dataset for this project is collected and annotated manually

Hardware

We have used Google Colab to train the model

| NVIDIA-SMI 455.45.01 Driver Version: 418.67 CUDA Version: 10.1 | | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | MIG M. |

| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 | | N/A 57C P8 10W / 70W | 0MiB / 15079MiB | 0% Default | ERR! |

Our Research

• Inorder to improve our detection and classification perfomance of the model we have performed some experiments on color augmentation parameters of YOLO V4 . We have adjusted the color augmentation parameters HUE, SATURATION and EXPOSURE and trained the YOLO V4 and analyzed the results .

• Checkout this to know more about the color augmentation parameters that we have experimented : Color Augmentation