relativeCameraPose icon indicating copy to clipboard operation
relativeCameraPose copied to clipboard

After I download DTU dataset, it can not run by the next guide.

Open ray2004a opened this issue 6 years ago • 1 comments

Thanks for the sharing, and sorry for the problem as a newer of Torch. I already downloaded full cleaned dataset on DTU(136GB) and extracted it to the "data" folder in roots of this git-master. Then run " th main.lua -do_evaluation -source_image_path data/Cleaned/ -weights ./pre-trained/siam_hybridnet_fullsized.t7", but it showed like below:

======================================================== invalid argument: -source_image_path Usage: [options]

Torch-7 relative pose estimation Options:

  • GPU Default preferred GPU [1]
  • nGPU Number of GPUs to use by default [2]
  • manualSeed Manually set RNG seed [333]
  • quaternions Use quaternion parametrization. Rodriguez otherwise [true]
  • spp Use spatial pyramid pooling [true]
  • do_evaluation Evaluation stage [false]
  • do_cropping Crop the original images [false]
  • data_description_path Home directory for text data [./data]
  • source_img_path Home directory for images of corresponding landmarks [./data/Cleaned]
  • image_size_x Image width [1200]
  • image_size_y Image width [1600]
  • crop_size Crop size [227]
  • training_dataset_size Number of training pairs [39424]
  • test_dataset_size Number of test pairs [35391]
  • learning_rate Learning rate [0.001]
  • gamma inv learning rate decay type: lr * (1 + gamma * epoch) ^ (-power) [0.001]
  • power inv learning rate decay type: lr * (1 + gamma * epoch) ^ (-power) [0.5]
  • weight_decay weight decay [1e-05]
  • beta1 first moment coefficient (for adam solver) [0.9]
  • beta2 second moment coefficient (for adam solver) [0.999]
  • train_batch_size Training mini-batch size [22]
  • test_batch_size Test (validation) mini-batch size [47]
  • epoch_number Manual epoch number (useful on restarts) [1]
  • max_epoch Number of training epochs [10]
  • snapshot_dir Snapshot directory [./snapshots]
  • results_dir Results directory [./results]
  • weights Pretrained model to begin training from [] ===============================================================

Could anyone face this problem before? It will be very thanked for any suggestion.

ray2004a avatar Jan 21 '19 10:01 ray2004a

Try to change the 19th line of opts.lua. E.g: cmd:option("-source_img_path", "/media/hahahehe/KESU/Cleaned", 'Home directory for images of corresponding landmarks') No need to add -source_img_path parameters when you run the command. th main.lua -do_evaluation -weights ./pre-trained/siam_hybridnet_fullsized.t7

hahahehe-coder avatar Mar 13 '21 06:03 hahahehe-coder