def save_anim(reference_positions, bone_dependencies, rotations, name, prep):
reference_anim = Animator.MocapAnimator2(reference_positions, [''] * 40,
bone_dependencies,
prep.target_delta_t,
heading_dirs=rotations,
name=name)
reference_anim.animation()
def save_anim(self, global_positions, reference_positions,
bone_dependencies, rotations, name):
file_name = name.split('/')[-1]
name = name.replace(file_name, "videos_out/" + file_name)
anim = Animator.MocapAnimator2(global_positions, [''] * 40,
bone_dependencies,
self.train_prep.target_delta_t,
heading_dirs=rotations,
name="trained.gif")
anim.animation()
reference_anim = Animator.MocapAnimator2(
reference_positions, [''] * 40,
bone_dependencies,
self.train_prep.target_delta_t,
heading_dirs=rotations,
name="reference.gif")
reference_anim.animation()
AnimationStacker.concatenate_animations("trained.gif", "reference.gif",
name + ".mp4")
# "l_hand_idx, r_hand_idx, l_elbow_idx, r_elbow_idx, hip_idx, l_foot_idx, r_foot_idx
#[l_hand_idx, r_hand_idx, l_shoulder_idx, r_shoulder_idx, hip_idx, l_foot_idx, r_foot_idx, l_elbow_idx, r_elbow_idx, l_knee_idx, r_knee_idx]
bone_dependencies = [[0, 7], [1, 8], [2, 4], [3, 4], [4, -1], [5, 9], [6, 10],
[7, 2], [8, 3], [9, 4], [10, 4]]
bone_dependencies = np.array(bone_dependencies)
global_positions = np.hstack(
(training_prep.scale_back_input(eval_input.detach().cpu().numpy())[:, :6],
training_prep.scale_back_output(target_output.detach().cpu().numpy())))
# global_positions = np.hstack((eval_input, eval_output))
global_positions = global_positions.reshape(global_positions.shape[0], -1, 3)
global_positions = eval_prep.add_heads(global_positions)
# training_prep.scale_back(global_positions)
if __name__ == '__main__':
anim = Animator.MocapAnimator(global_positions, [''] * 40,
bone_dependencies, 1.0 / TARGET_FPS)
anim.animation()
# Test the model
# In test phase, we don't need to compute gradients (for memory efficiency)
# with torch.no_grad():
# n_correct = 0
# n_samples = 0
# for input, output in test_loader:
# input = input.reshape(-1, 28 * 28).to(device)
# output = output.to(device)
# outputs = model(input)
# # max returns (value ,index)
# predicted = torch.round(10 * outputs.data).T
# n_samples += output.size(0)
# n_correct += (predicted == output).sum().item()
eval_output = eval_output[idx1:idx2, :]
bone_dependencies, global_positions, rotations = eval_prep.get_global_pos_from_prediction(
eval_input,
to_numpy(final_outputs),
training_prep,
start_idx=idx1,
end_idx=idx1 + 499)
bone_dependencies, reference_positions, rotations = eval_prep.get_global_pos_from_prediction(
eval_input, eval_output, training_prep, start_idx=idx1, end_idx=idx2)
MOTIONTYPE = "Boxing"
# sp.print_stats(global_positions, reference_positions, ["l_hand", "r_hand", "l_shoulder", "r_shoulder", "hip", "l_foot", "r_foot", "l_elbow", "r_elbow", "l_knee", "r_knee"], MOTIONTYPE)
if __name__ == '__main__':
anim = Animator.MocapAnimator(global_positions, [''] * 40,
bone_dependencies,
1.0 / TARGET_FPS,
heading_dirs=rotations,
name="trained.avi")
anim.animation()
reference_anim = Animator.MocapAnimator(reference_positions, [''] * 40,
bone_dependencies,
1.0 / TARGET_FPS,
heading_dirs=rotations,
name="reference.avi")
reference_anim.animation()
from Helpers import AnimationStacker
AnimationStacker.concatenate_animations("trained.avi", "reference.avi",
MOTIONTYPE + ".mp4")
#[l_hand_idx, r_hand_idx, l_shoulder_idx, r_shoulder_idx, hip_idx, l_foot_idx, r_foot_idx, l_elbow_idx, r_elbow_idx, l_knee_idx, r_knee_idx]
bone_dependencies = [[0, 7], [1, 8], [2, 4], [3, 4], [4, -1], [5, 9], [6, 10],
[7, 2], [8, 3], [9, 4], [10, 4]]
bone_dependencies = np.array(bone_dependencies)
global_positions = np.hstack(
(training_prep.scale_back_input(eval_input.detach().cpu().numpy())[:, :6],
training_prep.scale_back_output(target_output.detach().cpu().numpy())))
# global_positions = np.hstack((eval_input, eval_output))
global_positions = global_positions.reshape(global_positions.shape[0], -1, 3)
global_positions = eval_prep.add_heads(global_positions)
# training_prep.scale_back(global_positions)
if __name__ == '__main__':
anim = Animator.MocapAnimator(global_positions, [''] * 40,
bone_dependencies,
1.0 / TARGET_FPS,
write_to_file=True)
anim.animation()
# Test the model
# In test phase, we don't need to compute gradients (for memory efficiency)
# with torch.no_grad():
# n_correct = 0
# n_samples = 0
# for input, output in test_loader:
# input = input.reshape(-1, 28 * 28).to(device)
# output = output.to(device)
# outputs = model(input)
# # max returns (value ,index)
# predicted = torch.round(10 * outputs.data).T
# n_samples += output.size(0)
bone_dependencies, global_positions, _ = eval_prep.get_global_pos_from_prediction(
eval_input, to_numpy(final_outputs), training_prep)
_, reference_positions, _ = eval_prep.get_global_pos_from_prediction(
eval_input, eval_output, training_prep)
MOTIONTYPE = "Boxing"
sp.print_stats(global_positions, reference_positions, [
"l_hand", "r_hand", "l_shoulder", "r_shoulder", "hip", "l_foot", "r_foot",
"l_elbow", "r_elbow", "l_knee", "r_knee"
], MOTIONTYPE)
if __name__ == '__main__':
anim = Animator.MocapAnimator(
global_positions, [''] * 40,
bone_dependencies,
1.0 / TARGET_FPS,
heading_dirs=eval_prep.heading_dirs[-global_positions.shape[0]:],
name="trained.avi")
anim.animation()
reference_anim = Animator.MocapAnimator(
reference_positions, [''] * 40,
bone_dependencies,
1.0 / TARGET_FPS,
heading_dirs=eval_prep.heading_dirs[-reference_positions.shape[0]:],
name="reference.avi")
reference_anim.animation()
from Helpers import AnimationStacker
AnimationStacker.concatenate_animations("trained.avi", "reference.avi",
MOTIONTYPE + ".mp4")