"You must at least output one value !" when running inference on pretrained model

[evanc400]

Thanks for your generosity and hard work maintaining the repo!

I am currently trying to run inference on the pretrained model using KITTI sequences 09 and 10. For my dispnet, I am using the dispnet_model_best.pth.tar which you provided in the "Available here" link under pretrained nets. I am a Mac User. When I try to run: python3 run_inference.py --pretrained pretrained_nets/.dispnet_model_best.pth.tar.icloud --dataset-dir odom_data/dataset/sequences --output-dir output I am met with an error message which reads "You must at least output one value !". I am confused what I am doing wrong exactly as I thought I gave all the arguments requested. Any suggestions would be much appreciated!

[ClementPinard]

This script lets you output disparity or depth. you need to specify it with the option --output-disp or --output_depth (no argument). If you select neither, the script has nothing to save and thus won't run

[evanc400]

Thanks for your quick response!

So I added that command to my arguments: python3 run_inference.py --pretrained pretrained_nets/.dispnet_model_best.pth.tar.icloud --dataset-dir odom_data/dataset/sequences --output-depth --output-dir output and I am now met with this error:

I'm unsure what is causing this exactly but I'm guessing it has to do with loading one of the arguments perhaps the dispnet (dispnet_model_best.pth.tar.icloud) I got from the google drive link. Do you have ideas?

[ClementPinard]

It took me some time but I just tested it on my side with a new pytorch install and it runs fine. According to this link https://discuss.pytorch.org/t/unpickling-stack-underflow/68181/2 , the problem might be related to a corrupt file. Have you tried re-downloading it ? As a side note, this file is NOT a compressed file, you just need to use it as is in the CLI without trying to untar it