Exposing number of detections in the training script, and adding 16bit image reading support

[ajinkya-kulkarni]

Hi, thanks again for this repo. I was wondering if exposing number of detections in the training script, and adding 16bit image reading support is possible. I can also contribute towards the 16bit part.

[dnth]

@ajinkya-kulkarni I've added support for configuring model in v0.1.2. Would you please try it out and let me know if it works for you?

from deimkit import Config, configure_model

conf = Config.from_model_name("deim_hgnetv2_s")

# Optional
conf = configure_model(
    config=conf, 
    num_queries=100,   # Optional, default is 300
    pretrained=True,   # Optional, default is True
    freeze_at=-1       # Optional, default is -1 (no freezing)
)

[ajinkya-kulkarni]

Hey, thanks, I will give it a shot.
Regarding the 16 bit support, I plan to imread unchanged, then divide by 65535, so my output image is [0,1]. Will there be any problem downstream, maybe during augmentations or anything else that might break things?

[dnth]

I guess you'll just have to try it out. Let me know :)