nvlabs / bongard-logo Goto Github PK

Hi,

I want to run through the baselines using two v100-GPUs with 32GB CUDA memory of each but it still calls the error ''CUDA out of memory'', I wonder how it happened since there are total 64GB CUDA memory? (I've changed the gpu number in run_[model_name]_model.sh to fit the situation.)

Thank you !

Best,
Yi Xu

When I run the "run_moco_model.sh" script. It results in the following error. It seems like "moco" model does not exist in the models directory. How can we fix this?

Traceback (most recent call last):
File "train_moco.py", line 300, in
main(config)
File "train_moco.py", line 120, in main
model = models.make(config['model'], **config['model_args'])
File "/content/Bongard-LOGO/Bongard-LOGO_Baselines/models/models.py", line 20, in make
model = modelsname
KeyError: 'moco'

Hi,

Thank you for your excellent work! In the formal print, I found that MAML results were not reported. But did exist MAML code and related scripts under the Bongard-LOGO_Baselines/.

I tried to run those scripts but got an error shown as below.

File "*/Bongard-LOGO_Baselines/models/maml.py", line 145, in forward
    return F.linear(input, params['weight'], bias)
KeyError: 'weight'

The params is an empty orderedDict as I print it out. I used both run_maml_model.sh and run_maml_v2_model.sh to train the model, but got the same error. Is there anything I can do to fix it?

Thank you for any help.

Is there a way to get access to pretrained encoder needed for Meta-Baseline-MoCo? (without training it from scratch)

python run.py throws an error when trying to run from Bongard-LOGO/examples/02-bongard_logo/

This happens 'cuz the bongard module is not accessible from this folder.
I fixed it by moving the 'bongard' folder to Bongard-LOGO/examples/02-bongard_logo/ but this might not be the best solution.
There might be a need to arrange the directories.

Thank you for your work. The data Bongard-LOGO zip file (ShapeBongard_V2.zip) has the folders bb, ff, hd, and I don't know what they correspond to in the paper. I suppose ff is freeform, but for the others I am lost.

When creating Bongard Problem by using import_from_action_string_list, why do get_start_coordinates() return list of None?
Even positive_bongard_image.get_start_coordinates() for positive_bongard_image in bongard_problem_recovered.positive_bongard_images return list of None

I am using Demo and then bongard_problem_recovered.get_start_coordinates()