Skip to content

Latest commit

 

History

History
44 lines (24 loc) · 1.51 KB

Data_Weight_Preparation.md

File metadata and controls

44 lines (24 loc) · 1.51 KB
Raw

Guidance for Preparing Datasets and Pretrained Weights

1. Download Datasets

You could consider which dataset you would like to train/val, and download several of them in the following links:

Download SBD, GrabCut, Berkeley, DAVIS, Pascal VOC, COCO, LVIS following ritm project page

DAVIS-585: CliXEG

MSRA10K: https://mmcheng.net/msra10k/

DUTS-TE+TR: http://saliencydetection.net/duts/

HFlicker: https://github.com/bcmi/Image-Harmonization-Dataset-iHarmony4

YoutubeVOS: https://competitions.codalab.org/competitions/20127#participate-get-data

ThinObject: https://github.com/liewjunhao/thin-object-selection

2. Download Imagenet Pretrained Weights

SegFormer : https://drive.google.com/drive/folders/1b7bwrInTW4VLEm27YawHOAMSMikga2Ia
HRNet : https://github.com/HRNet/HRNet-Image-Classification

3. Set Paths in Config

Do not forget to edit the path to datasets and pretrained weights in ./config.yml

Adding your own dataset

If you want to train/val the model on your own dataset.

  1. Find a templet in ./isegm/data/datasets/ and write your own dataset.
  2. Add the dataset in ./isegm/data/datasets/_init_.py
  3. To val on the dataset: edit the function 'get_dataset' in ./isegm/inference/utils.py; add the dataset name in ./trainval_scripts/val_xxxx.sh.
  4. To train on the dataset: add the dataset in the corresponding files in ./models.