2.3 KiB

Raw Blame History

Prepare Datasets

[Prepare Cifar](#Prepare Cifar)
[Prepare Imagenet](#Prepare Imagenet)
[Prepare Imagenet-TFrecords](#Prepare Imagenet-TFrecords)
[Prepare COCO](#Prepare COCO)
[Prepare PAI-Itag detection](#Prepare PAI-Itag detection)

Prepare Cifar

Download dataset cifar10 and uncompress files to data/cifar, directory structure is as follows:

data/cifar
└── cifar-10-batches-py
    ├── batches.meta
    ├── data_batch_1
    ├── data_batch_2
    ├── data_batch_3
    ├── data_batch_4
    ├── data_batch_5
    ├── readme.html
    ├── read.py
    └── test_batch

Prepare Imagenet

Go to the download-url, Register an account and log in .
Download the following files：
- Training images (Task 1 & 2). 138GB.
- Validation images (all tasks). 6.3GB.
Unzip the downloaded file.
Using this scrip to get data meta.

Prepare Imagenet-TFrecords

Go to the download-url, Register an account and log in .
The dataset is divided into two parts, part0 (79GB) and part1 (75GB), you need download all of them.

Prepare COCO

Download COCO2017 dataset to data/coco, directory structure is as follows

data/coco
├── annotations
├── train2017
└── val2017

Prepare PAI-Itag detection

Download SmallCOCO dataset to data/coco, directory structure is as follows:

data/coco/
├── train2017
├── train2017_20_local.manifest
├── val2017
└── val2017_20_local.manifest

replace train_data and val_data path in config file

sed -i 's#train2017.manifest#train2017_20_local.manifest#g' configs/detection/yolox_coco_pai.py
sed -i 's#val2017.manifest#val2017_20_local.manifest#g' configs/detection/yolox_coco_pai.py

2.3 KiB Raw Blame History Unescape Escape