I have some problems preparing the dataset.
Closed this issue · 4 comments
Suncheng2022 commented
Hi Dear Author, I have some problems preparing the dataset:
- I am preparing the COCO dataset according to your project directory, but there is no "instances_valminusminival2014.json" file in the origin COCO2014 or COCO2017 dataset. Is this file generated by yourself, and what should I do to generate it?
- About the LVIS data set, I wonder if your project has preprocessed it when using the LVIS dataset, can I directly go to the lvis official website to download the v1.0 version (because it does not currently provide the v0.5 version) modify the train script and use it directly ?
Thanks in advance!
ruifenggong commented
Hi Dear Author, I have some problems preparing the dataset:
- I am preparing the COCO dataset according to your project directory, but there is no "instances_valminusminival2014.json" file in the origin COCO2014 or COCO2017 dataset. Is this file generated by yourself, and what should I do to generate it?
- About the LVIS data set, I wonder if your project has preprocessed it when using the LVIS dataset, can I directly go to the lvis official website to download the v1.0 version (because it does not currently provide the v0.5 version) modify the train script and use it directly ?
Thanks in advance!
去这里下一个annotations_trainval2017.zip,解压过去就有了
https://aistudio.baidu.com/aistudio/datasetdetail/7122
Suncheng2022 commented
- instances_valminusminival2014.json
没有这个文件—— instances_valminusminival2014.json
zhenyuw16 commented
instances_valminusminival2014 is actually images that belong to train2017 but are not in train2014. You can create it in this way if you cannot find the annotation file. You can also randomly extract about 35k images from train2017 directly, and the performance will not vary too much.
Suncheng2022 commented
I just search instances_valminusminival2014.json on Internet and got it in here