How to configure for variable image dimension ?

[nikhilparmar]

What changes do I need to make so that I can train with variable image dimensions ? What strategy do I need to follow to solve this issue ?

Apologies for such questions, Thanks

[michhar]

Hi, Thanks for the good question. I have not tried with variable image sizes. Could you provide the error you see when you try? I will try to help if I can.

You may wish to look in the config.py at:

• https://github.com/michhar/maskrcnn-custom/blob/aad9724db054069fa49bb11a2775f550bd4034d9/mrcnn/config.py#L211
• https://github.com/michhar/maskrcnn-custom/blob/aad9724db054069fa49bb11a2775f550bd4034d9/custom.py#L95

Thanks.

[nikhilparmar]

Hi, Thanks for the good question. I have not tried with variable image sizes. Could you provide the error you see when you try? I will try to help if I can.

You may wish to look in the config.py at:

• https://github.com/michhar/maskrcnn-custom/blob/aad9724db054069fa49bb11a2775f550bd4034d9/mrcnn/config.py#L211
• https://github.com/michhar/maskrcnn-custom/blob/aad9724db054069fa49bb11a2775f550bd4034d9/custom.py#L95

Thanks.

Okay nice. So if at line no. 95, if I make it 1024x1024, then all the images will be resized to it right ? Also will it make necessary changes to the annotation coordinates as per the mentioned dimensions ?

Thanks

[michhar]

@nikhilparmar I'm afraid I'm not sure of the downstream effects as I haven't tried, but did this help you solve the variable image dimension issue? Thanks!