search

rozumden / FMODetect

[ICCV 2021] FMODetect: Robust Detection of Fast Moving Objects
https://github.com/rozumden/FMODetect
MIT License
44 stars 5 forks source link

How to train custom fast moving object detction #5

Open Manueljohnson063 opened 2 months ago

Manueljohnson063 commented 2 months ago

Hai,First of all thanks for your great repo... I am planning to train your FMO code in my dataset . Is it possible ? The data ie is available with me is only images captured from camera @rozumden

rozumden commented 2 months ago

Hi, there is a training script: https://github.com/rozumden/FMODetect/blob/master/train.py, which can be triggered with train.sh. The synthetic dataset is generated using this sub-folder: https://github.com/rozumden/FMODetect/tree/master/dataset Since you have only images captured with your camera, I assume you must provide ground truth by manual labelling.

Manueljohnson063 commented 2 months ago

Hi sir , How i will annotate my dataset. Specifically any tools available ? Many many thanks in advance. @rozumden

Manueljohnson063 commented 2 months ago

Hi ,Sir I am having images that have FMO ,Which tool you had used to annotate the custom data .Please help me Thanks in advance . @rozumden

rozumden commented 2 months ago

Hi, we used synthetic data to train, and we used Blender to generate it. For your use-case, you could just annotate with a custom written annotation tool in Python.

Manueljohnson063 commented 2 months ago

Sir,Thanks for your quick response . Is it same like annotating the object detection ? How i select the area? How you annotate your blender dataset ? Sir... @rozumden