Skip to content

Latest commit

 

History

History

generate_mxnet_dataset

README.md

Generate MXNet Use Dataset

Recent Update

2021-05-19: Fix generate pairs.txt with generate_pairs.py. Use 10-fold cross-validation. See here.

2021-05-06: First commit.

Content

Generate MXNet Use Dataset

Set Up

Python 3.6

CPU Version

Package Version
mxnet 1.7.0.post2
numpy 1.19.5
esaydict 1.9
psutil 5.8.0

GPU Version (CUDA 10.1)

Package Version
mxnet-cu101 1.5.0
numpy 1.19.5
esaydict 1.9
psutil 5.8.0

Create Python

conda create -n Python3.6 python=3.6

Install Python Package

CPU Version

requirements.txt

pip install -r requirements.txt

GPU Version (CUDA 10)

requirements-gpu.txt

pip install -r requirements-gpu.txt

Generate .rec

usage: generate_rec.py [-h] [--datasets_images_dir DATASETS_IMAGES_DIR]
                       [--output_path OUTPUT_PATH] [--image_size IMAGE_SIZE]
                       [--image_ext IMAGE_EXT] [--color {-1,0,1}]

Create an image list or make a record database by reading from an image list

optional arguments:
  -h, --help            show this help message and exit
  --datasets_images_dir DATASETS_IMAGES_DIR
                        is your dataset images directory. 
                        (default:../../../../Deep Learning/InsightFace/Python/Dataset/faces_emore_mask/images)
  --output_path OUTPUT_PATH
                        is where you want to save .rec 
                        (default:../../../../Deep Learning/InsightFace/Python/Dataset/faces_emore_mask/train.rec)
  --image_size IMAGE_SIZE
                        is dataset image size. 
                        (default: 112,112)
  --image_ext IMAGE_EXT
                        is dataset images extension. 
                        (default: jpg)
  --color {-1,0,1}      specify the color mode of the loaded image. 
                        1: Loads a color image. Any transparency of image will be neglected. It is the default flag. 
                        0: Loads image in grayscale mode. 
                        -1:Loads image as such including alpha channel. (default: 1)

python generate_rec.py --datasets_images_dir "../../../../Deep_Learning/InsightFace/Python/Dataset/faces_emore_mask/images" --output_path "../../../../Deep_Learning/InsightFace/Python/Dataset/faces_emore_mask/train.rec"

Generate pairs.txt

usage: generate_pairs.py [-h] [--datasets_images_dir DATASETS_IMAGES_DIR] [--output_path OUTPUT_PATH]
                         [--pairs_num PAIRS_NUM] [--image_ext IMAGE_EXT]

Generate pairs.txt

optional arguments:
  -h, --help            show this help message and exit
  --datasets_images_dir DATASETS_IMAGES_DIR
                        is your dataset images directory.
  --output_path OUTPUT_PATH
                        is where you want to save pairs.txt.
  --pairs_num PAIRS_NUM
                        is how many pairs that you want to create.
  --image_ext IMAGE_EXT
                        is dataset images extension.

python generate_pairs.py --datasets_images_dir "../../../../Deep_Learning/InsightFace/Python/Dataset/faces_emore_mask/images" --output_path "../../../../Deep_Learning/InsightFace/Python/Dataset/faces_emore_mask/pairs.txt" --pairs_num 3000 --image_ext jpg

Generate .bin

usage: generate_bin.py [-h] [--pairs_txt_path PAIRS_TXT_PATH]
                       [--output_bin_path OUTPUT_BIN_PATH]

Generate pin

optional arguments:
  -h, --help            show this help message and exit````
  --pairs_txt_path PAIRS_TXT_PATH
                        where is the pairs.txt.
  --output_bin_path OUTPUT_BIN_PATH
                        where to save the output .bin.

python generate_bin.py --pairs_txt_path "../../../../Deep_Learning/InsightFace/Python/Dataset/faces_emore_mask/pairs.txt" --output_bin_path "../../../../Deep_Learning/InsightFace/Python/Dataset/faces_emore_mask/faces_emore_mask.bin"