execution time limited

Hey, I was wondering if the execution times of the containers is limited again? My submission fails also if it takes equal amount of time to execute as other submissions that suceeded. Thanks in advance

Hi,

Your algorithm can run a maximum of 1 hour to process 100 CXR images. (that includes all the steps such as creating the model, loading the weights and inference)

I just asked the grand-challenge team to verify if that is the case. I will let you know.

Best, Ecem

Thanks, Could you tell me if the failed algorithm is using the GPU correctly? it should print "using devide: cuda" somewhere. If not this might be a problem.

best Finn

Hi,

The only print is this:

2022-01-21T07:31:40+00:00 parameters are

So, i think it does not use the GPU. BTW, i have verifed that the inference time is indeed a maximum of 1 hour for 100 CXR images.

Best, Ecem

Hey,

My algos keep failing due to runtime i guess. However, the inference time on my pc is about 8mins for 350 images. could you have a look at my last successfull submission (Test7 Algo)? there should be logs and I would like to see if the algorithm uses the GPU. I printed print('using the device ', self.device) At init. Could you provide me with details about the used device? Thanks in advance, Finn

Hi,

Your successful submission uses CPU:

2022-01-23T17:38:22+00:00 parameters are 2022-01-23T18:08:53+00:00 using the device cpu 2022-01-23T18:08:53+00:00 1.10.1+cu102 2022-01-23T18:08:53+00:00 LOADING last Checkpoints for fcrnn_l 2022-01-23T18:08:53+00:00 LOADING last Checkpoints for retina_l 2022-01-23T18:08:53+00:00 LOADING last Checkpoints for effdet2_l 2022-01-23T18:08:53+00:00 FAILED TO LOAD STATE DICT - THIS IS OK IN EVALUATION 2022-01-23T18:08:53+00:00 USING E 2022-01-23T18:08:53+00:00 FAILED TO LOAD STATE DICT - THIS IS OK IN EVALUATION 2022-01-23T18:08:53+00:00 USING E 2022-01-23T18:08:53+00:00 FAILED TO LOAD STATE DICT - THIS IS OK IN EVALUATION 2022-01-23T18:08:53+00:00 USING E 2022-01-23T18:08:53+00:00 FAILED TO LOAD STATE DICT - THIS IS OK IN EVALUATION 2022-01-23T18:08:53+00:00 USING E 2022-01-23T18:08:53+00:00 FAILED TO LOAD STATE DICT - THIS IS OK IN EVALUATION 2022-01-23T18:08:53+00:00 USING E 2022-01-23T18:08:53+00:00 LOADING best_k Checkpoints for yolo_s 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 LOADING best_k Checkpoints for yolo_l 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 USING Y 2022-01-23T18:08:53+00:00 done 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred 2022-01-23T18:08:53+00:00 pred

The failed submission only prints this: 2022-01-24T10:12:48+00:00 parameters are

Did you configure your algorithm to use GPU in grand-challenge?

there is a checkbox written Image requires gpu which you need to select?

Best, Ecem

Yes i did configure that checkbox. I will try to submit another container with another pytorch/cuda version and hope for the best.. BTW. im using this command self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu") for the device. And the labels from the Dockerfile are : LABEL nl.diagnijmegen.rse.algorithm.hardware.cpu.count=2 LABEL nl.diagnijmegen.rse.algorithm.hardware.cpu.capabilities=() LABEL nl.diagnijmegen.rse.algorithm.hardware.memory=12G LABEL nl.diagnijmegen.rse.algorithm.hardware.gpu.count=1 LABEL nl.diagnijmegen.rse.algorithm.hardware.gpu.cuda_compute_capability= LABEL nl.diagnijmegen.rse.algorithm.hardware.gpu.memory=10G just as in your example. im wondering why "cuda_compute_capability= "has no value. Is this normal?

Best, Finn