When running simulate.py as described in the README, I'm encountering the following error:
Traceback (most recent call last): File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\simulate.py", line 350, in <module> simulate(planner_cfg, agent_cfg, filter_cfg, extra_cfg, density_fn, render_fn, get_rays_fn) File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\simulate.py", line 88, in simulate state_est = filter.estimate_state(gt_img, true_pose, action) File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\nav\estimator_helpers.py", line 374, in estimate_state xt, success_flag = self.estimate_relative_pose(sensor_img, self.xt.clone().detach(), sig_prop, obs_img_pose=obs_img_pose) File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\nav\estimator_helpers.py", line 240, in estimate_relative_pose loss.backward() File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\_tensor.py", line 492, in backward torch.autograd.backward( File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\autograd\__init__.py", line 251, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\autograd\function.py", line 288, in apply return user_fn(self, *args) File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\autograd\function.py", line 404, in backward raise NotImplementedError( NotImplementedError: You must implement either the backward or vjp method for your custom autograd.Function to use it with backward mode AD.
For whatever reason, pytorch seems to be complaining about this loss function. Has anyone else come across this before and found a solution? (For reference I'm using pytorch 2.1.0 + CUDA 11.8).
When running simulate.py as described in the README, I'm encountering the following error:
Traceback (most recent call last): File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\simulate.py", line 350, in <module> simulate(planner_cfg, agent_cfg, filter_cfg, extra_cfg, density_fn, render_fn, get_rays_fn) File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\simulate.py", line 88, in simulate state_est = filter.estimate_state(gt_img, true_pose, action) File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\nav\estimator_helpers.py", line 374, in estimate_state xt, success_flag = self.estimate_relative_pose(sensor_img, self.xt.clone().detach(), sig_prop, obs_img_pose=obs_img_pose) File "C:\Users\jaf25\OneDrive\Documents\Github\nerf-navigation\nav\estimator_helpers.py", line 240, in estimate_relative_pose loss.backward() File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\_tensor.py", line 492, in backward torch.autograd.backward( File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\autograd\__init__.py", line 251, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\autograd\function.py", line 288, in apply return user_fn(self, *args) File "C:\Users\jaf25\anaconda3\lib\site-packages\torch\autograd\function.py", line 404, in backward raise NotImplementedError( NotImplementedError: You must implement either the backward or vjp method for your custom autograd.Function to use it with backward mode AD.For whatever reason, pytorch seems to be complaining about this loss function. Has anyone else come across this before and found a solution? (For reference I'm using pytorch 2.1.0 + CUDA 11.8).