ImportError: Cannot load backend 'TkAgg' which requires the 'tk' interactive framework, as 'headless' is currently running

[zy-lazytitan]

I've configured the environment, then ran the code python run_cf3dgs.py -s data/Tanks/Francis/ --mode train --data_type tanks. It reported an error:

Traceback (most recent call last):
File "/home/l/disk_4T/zz/CF-3DGS/run_cf3dgs.py", line 13, in
from trainer.cf3dgs_trainer import CFGaussianTrainer
File "/home/l/disk_4T/zz/CF-3DGS/trainer/cf3dgs_trainer.py", line 51, in
from utils.vis_utils import interp_poses_bspline, generate_spiral_nerf, plot_pose
File "/home/l/disk_4T/zz/CF-3DGS/utils/vis_utils.py", line 16, in
from evo.tools import plot
File "/home/l/anaconda3/envs/cf3dgs/lib/python3.10/site-packages/evo/tools/plot.py", line 83, in
apply_settings(SETTINGS)
File "/home/l/anaconda3/envs/cf3dgs/lib/python3.10/site-packages/evo/tools/plot.py", line 63, in apply_settings
mpl.use(settings.plot_backend)
File "/home/l/anaconda3/envs/cf3dgs/lib/python3.10/site-packages/matplotlib/init.py", line 1233, in use
plt.switch_backend(name)
File "/home/l/anaconda3/envs/cf3dgs/lib/python3.10/site-packages/matplotlib/pyplot.py", line 279, in switch_backend
raise ImportError(
ImportError: Cannot load backend 'TkAgg' which requires the 'tk' interactive framework, as 'headless' is currently running

Is this because the version of "evo" is wrong? My evo version is 1.29.0 and the matplotlib version is 3.7.0, as mentioned in requirement.txt

[pdd-show]

I ran into the same problem, saying 'ImportError: Cannot load backend 'TkAgg' which requires the 'tk' interactive framework, as 'headless' is currently running ',
My ubuntu is Ubuntu 22.04 LTS server, other issues say that the matplotlib version is too high, I reduced to version 3.7, but it did not work.
Or add 'matplotlib.use('Agg')' to the py file, which doesn't work either.

[massyzs]

Fixed:
Please remove all evo related lib and only rely on matplotlib. And then:
Use code generated by GPT: (I just get tired of configuring this env.)

def align_trajectories(ref_poses, est_poses):
# Example alignment logic (dummy, for actual use, replace with proper transformation logic)
# This assumes both are numpy arrays of shape (N, 3)
est_poses_aligned = copy.deepcopy(est_poses)
# Assuming simple translation alignment (you can adjust this logic as needed)
offset = np.mean(ref_poses - est_poses, axis=0)
est_poses_aligned += offset
return est_poses_aligned

def plot_pose(ref_poses, est_poses, output_path, vid=False):
ref_poses = np.array(ref_poses)
est_poses = np.array(est_poses)

# Align trajectories
traj_ref = ref_poses
traj_est_aligned = align_trajectories(ref_poses, est_poses)

if vid:
    for p_idx in range(len(ref_poses)):
        fig = plt.figure()
        ax = fig.add_subplot(111, projection="3d")

        current_est_aligned = traj_est_aligned[:p_idx + 1]
        current_ref = traj_ref[:p_idx + 1]

        ax.plot(current_ref[:, 0], current_ref[:, 1], current_ref[:, 2], 'b--', label="Ground-truth")
        ax.plot(current_est_aligned[:, 0], current_est_aligned[:, 1], current_est_aligned[:, 2], 'r-', label="Ours (aligned)")

        ax.set_xlabel('X')
        ax.set_ylabel('Y')
        ax.set_zlabel('Z')
        ax.legend()
        ax.view_init(elev=10., azim=45)

        os.makedirs(os.path.join(os.path.dirname(output_path), 'pose_vid'), exist_ok=True)
        pose_vis_path = os.path.join(os.path.dirname(output_path), 'pose_vid', f'pose_vis_{p_idx:03d}.png')
        print(pose_vis_path)
        fig.savefig(pose_vis_path)
        plt.close(fig)

# Static plot
fig = plt.figure()
ax = fig.add_subplot(111, projection="3d")

ax.plot(traj_ref[:, 0], traj_ref[:, 1], traj_ref[:, 2], 'b--', label="Ground-truth")
ax.plot(traj_est_aligned[:, 0], traj_est_aligned[:, 1], traj_est_aligned[:, 2], 'r-', label="Ours (aligned)")

ax.set_xlabel('X')
ax.set_ylabel('Y')
ax.set_zlabel('Z')
ax.legend()
ax.view_init(elev=10., azim=45)

pose_vis_path = os.path.join(os.path.dirname(output_path), 'pose_vis.png')
fig.savefig(pose_vis_path)
plt.close(fig)

[pdd-show]

Fixed: Please remove all evo related lib and only rely on matplotlib. And then: Use code generated by GPT: (I just get tired of configuring this env.)

def align_trajectories(ref_poses, est_poses): # Example alignment logic (dummy, for actual use, replace with proper transformation logic) # This assumes both are numpy arrays of shape (N, 3) est_poses_aligned = copy.deepcopy(est_poses) # Assuming simple translation alignment (you can adjust this logic as needed) offset = np.mean(ref_poses - est_poses, axis=0) est_poses_aligned += offset return est_poses_aligned

def plot_pose(ref_poses, est_poses, output_path, vid=False): ref_poses = np.array(ref_poses) est_poses = np.array(est_poses)

# Align trajectories
traj_ref = ref_poses
traj_est_aligned = align_trajectories(ref_poses, est_poses)

if vid:
    for p_idx in range(len(ref_poses)):
        fig = plt.figure()
        ax = fig.add_subplot(111, projection="3d")

        current_est_aligned = traj_est_aligned[:p_idx + 1]
        current_ref = traj_ref[:p_idx + 1]

        ax.plot(current_ref[:, 0], current_ref[:, 1], current_ref[:, 2], 'b--', label="Ground-truth")
        ax.plot(current_est_aligned[:, 0], current_est_aligned[:, 1], current_est_aligned[:, 2], 'r-', label="Ours (aligned)")

        ax.set_xlabel('X')
        ax.set_ylabel('Y')
        ax.set_zlabel('Z')
        ax.legend()
        ax.view_init(elev=10., azim=45)

        os.makedirs(os.path.join(os.path.dirname(output_path), 'pose_vid'), exist_ok=True)
        pose_vis_path = os.path.join(os.path.dirname(output_path), 'pose_vid', f'pose_vis_{p_idx:03d}.png')
        print(pose_vis_path)
        fig.savefig(pose_vis_path)
        plt.close(fig)

# Static plot
fig = plt.figure()
ax = fig.add_subplot(111, projection="3d")

ax.plot(traj_ref[:, 0], traj_ref[:, 1], traj_ref[:, 2], 'b--', label="Ground-truth")
ax.plot(traj_est_aligned[:, 0], traj_est_aligned[:, 1], traj_est_aligned[:, 2], 'r-', label="Ours (aligned)")

ax.set_xlabel('X')
ax.set_ylabel('Y')
ax.set_zlabel('Z')
ax.legend()
ax.view_init(elev=10., azim=45)

pose_vis_path = os.path.join(os.path.dirname(output_path), 'pose_vis.png')
fig.savefig(pose_vis_path)
plt.close(fig)

It worked. Thank you very much!