How to train Custom Dataset

This guide explains how to train your own custom dataset with fastreid’s data loaders.

Before You Start

Following Getting Started to setup the environment and install requirements.txt dependencies.

Train on Custom Dataset

1. Register your dataset (i.e., tell fastreid how to obtain your dataset).

   To let fastreid know how to obtain a dataset named “my_dataset”, users need to implement a Class that inherits fastreid.data.datasets.bases.ImageDataset:

   from fastreid.data.datasets import DATASET_REGISTRY
   from fastreid.data.datasets.bases import ImageDataset
   
   
   @DATASET_REGISTRY.register()
   class MyOwnDataset(ImageDataset):
   	def __init__(self, root='datasets', **kwargs):
   		...
   		super().__init__(train, query, gallery)		
   

   Here, the snippet associates a dataset named “MyOwnDataset” with a class that processes train set, query set and gallery set and then pass to the baseClass. Then add a decorator to this class for registration.

   The class can do arbitrary things and should generate train list: list(str, str, str), query list: list(str, int, int) and gallery list: list(str, int, int) as below.

   train_list = [
   (train_path1, pid1, camid1), (train_path2, pid2, camid2), ...]
   
   query_list = [
   (query_path1, pid1, camid1), (query_path2, pid2, camid2), ...]
   
   gallery_list = [
   (gallery_path1, pid1, camid1), (gallery_path2, pid2, camid2), ...]
   

   You can also pass an empty train_list to generate a “Testset” only with super().__init__([], query, gallery).

   Notice: query and gallery sets could have the same camera views, but for each individual query identity, his/her gallery samples from the same camera are excluded. So if your dataset has no camera annotations, you can set all query identities camera number to 0 and all gallery identities camera number to 1, then you can get the testing results.

2. Import your dataset.

   Aftre registering your own dataset, you need to import it in train_net.py to make it effective.

   from dataset_file import MyOwnDataset
   

[L1aoXingyu]

@AnhPC03 Yes, you are right! It doesn’t matter how your data structure is. The key idea is preparing the train, query and gallery as required and then pass via super().__init__(train, query, gallery).

[github-actions[bot]]

This issue was closed because it has been inactive for 14 days since being marked as stale.