JialeCao001 / SipMask

SipMask: Spatial Information Preservation for Fast Image and Video Instance Segmentation (ECCV2020)
https://arxiv.org/pdf/2007.14772.pdf
MIT License
334 stars 54 forks source link

Annotation format for a huge Dataset #57

Closed Darshcg closed 3 years ago

Darshcg commented 3 years ago

Hi @JialeCao001,

Thanks for your Amazing work!.

I have a Dataset consisting of 8Million images and I have a separate annotation file for each image which is in JSON format. The sample JSON is available at this link: https://drive.google.com/file/d/12lASRiHGm5fhbHP5YNCaQM-SwfvnGIcR/view?usp=sharing. This is for only one image, similarly, we have it for all the images.

My Query here is: do we have an annotation format having one annotation file per image that is supported by the repo? If yes, I will then write a simple script for the format conversion.

But the issue with using the Coco format(single format file for all the images) is that: The single JSON file for 8Million images is becoming very big and it eats up the memory.

So, what other alternative way I can look for with the present huge Dataset and the annotations for training the model?

looking forward to your reply.

Thanks and Regards