The repository provides code for running inference with the Meta Segment Anything Model 2 (SAM 2), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.
Apache License 2.0
12.14k
stars
1.1k
forks
source link
Question about max_num_objects parameter in training config #409
I am currently working on finetuning the model on my own dataset. However, I am having some trouble with the max_num_objects parameter in the config file. The more I increase this parameter, the higher are my compute requirements. My images have 1-50 objects each and I already struggle with setting the value to 20.
So, I have two questions:
What does the parameter do? Will this randomly choose a maximum of max_num_objects for each image? If so, will my model also converge, if the parameter is set lower, let´s say 5-10?
Why does this increase the compute requirements? Are the prediction processes for all single objects handled parallelly? If so, can this be changed to work in sequence, even if this would increase the training time?
I would greatly appreciate an answer to this, as it currently restricts me in continuing with my research project.
I am currently working on finetuning the model on my own dataset. However, I am having some trouble with the
max_num_objectsparameter in the config file. The more I increase this parameter, the higher are my compute requirements. My images have 1-50 objects each and I already struggle with setting the value to 20.So, I have two questions:
max_num_objectsfor each image? If so, will my model also converge, if the parameter is set lower, let´s say 5-10?I would greatly appreciate an answer to this, as it currently restricts me in continuing with my research project.
Thank you in advance!