search

lmb-freiburg / netdef_models

Repository for different network models related to flow/disparity (ECCV 18)
GNU General Public License v3.0
157 stars 32 forks source link

Error on "python3 controller.py " in occ-fill #5

Closed vhik4596 closed 5 years ago

vhik4596 commented 5 years ago

I was trying to running the script below in SceneFlow/occ-fill folder: python3 controller.py eval --imgs_t0 test_images/t0_imgL.png test_images/t0_imgR.png --imgs_t1 test_images/t1_imgL.png test_images/t1_imgR.png --dn_path ../../DispNet3/CSS/ --fn_path ../../FlowNet3/CSS/ But I got following error message: FileNotFoundError: [Errno 2] No such file or directory: '../../DispNet3/CSS/training/00__flyingThings3D.train__S_custom/checkpoints' I sure there is a folder in my computer named ../../DispNet3/CSS/training/00__flyingThings3D.train__S_short/checkpoints' after bash download_snapshots.sh Any solution? thanks.

tonmoy-saikia commented 5 years ago

Have you downloaded the snapshots for FlowNet3CSS and DispNet3CSS?

vhik4596 commented 5 years ago

Have you downloaded the snapshots for FlowNet3CSS and DispNet3CSS?

Yes, I have downloaded all snapshots by bash download_snapshots.sh in DispNet3 and FlowNet3 folder, respectively.

tonmoy-saikia commented 5 years ago

Looks like it is looking for the a wrong checkpoint directory.. train__S_custom vs train__S_short. You could manually try renaming it and see it works?

vhik4596 commented 5 years ago

Well, after I manually rename DispNet3/CSS/training/00__flyingThings3D.train__S_short to DispNet3/CSS/training/00__flyingThings3D.train__S_custom, there is a new error message: FileNotFoundError: [Errno 2] No such file or directory: '../../FlowNet3/CSS/training/01__flyingThings3D.train__S_short/checkpoints'

then, I tried rename FlowNet3/CSS/training/00__flyingThings3D.train__S_fine_half to FlowNet3/CSS/training/01__flyingThings3D.train__S_short/checkpoints, another error message happened:

.
.
.
restoring     refine_disp/encoder/conv6_1/kernel
restoring     refine_disp/encoder/predict/conv/bias
restoring     refine_disp/encoder/predict/conv/kernel
Saving output in: None
Traceback (most recent call last):
  File "controller.py", line 128, in <module>
    controller.run()
  File "/home/mark/.local/lib/python3.5/site-packages/netdef_slim/tensorflow/controller/base_controller.py", line 36, in run
    self._command_hooks[command]()
  File "controller.py", line 121, in eval
    state=self._args.state)
  File "/home/mark/.local/lib/python3.5/site-packages/netdef_slim/tensorflow/controller/base_controller.py", line 96, in eval
    out_path = os.path.join(out_dir, k+suffix)
  File "/usr/lib/python3.5/posixpath.py", line 89, in join
    genericpath._check_arg_types('join', a, *p)
  File "/usr/lib/python3.5/genericpath.py", line 143, in _check_arg_types
    (funcname, s.__class__.__name__)) from None
TypeError: join() argument must be str or bytes, not 'NoneType'
vhik4596 commented 5 years ago

Hi, @tonmoy-saikia I tried to running the script by adding --out_dir ./ (i.e. python3 controller.py eval --imgs_t0 test_images/t0 _imgL.png test_images/t0_imgR.png --imgs_t1 test_images/t1_imgL.png test_images/t1_imgR.png --dn_path ../../DispNet3/CSS/ --fn_path ../../FlowNet3/CSS/ --out_dir ./) Then, the program works! But, I am not sure manually rename DispNet3/CSS/training/00__flyingThings3D.train__S_short to DispNet3/CSS/training/00__flyingThings3D.train__S_custom and FlowNet3/CSS/training/00__flyingThings3D.train__S_fine_half to FlowNet3/CSS/training/01__flyingThings3D.train__S_short affect the result or not in my computer. Is there any groundtruth or reference that can let me check? BTW, this is the output and the output meaning "predicted disp from t=1 at t=0" in the ECCV paper Fig. 1.(d), is this right?

tonmoy-saikia commented 5 years ago

The output looks fine to me. You can use the KITTI or FlyingThings3D dataset to compare against the ground truth.

vhik4596 commented 5 years ago

@tonmoy-saikia Thanks!