This is really amazing! Can you tell me exactly how to do it? I really want to learn. I have also watched the teaching of binding skirts you did before, and it is really very good. Thank you very much.
Hi, It's really amazing. But I'm confused about the part of gathering data and trainning network. Does the data only contain information about skirt joint? Does neural network work this way:input the action of the character's bones and predict the joint information of the skirt?Can you tell me more information about these. Thank you very much.
Hi! Inputs are leg joints' X axis (X axis for upper/lower legs, 3*4 doubles), Outputs are skirt joints translations/rotations. What's confusing? I use ncloth with skirt joints attached to it to train the network. I rotate legs by a lot of angles and calculate ncloth dynamics with my skirt joints. So I have a lot of X axes for legs and skirt joints in each position. After I feed custom X axes and get desirable skirt joints' translation/rotation.
@@azagoruyko Hi,After your explanation, I tried to implement it these days, But I have some troubles on gathering data.Can you help me? Here's my problem. I made a skirt by myself and added joints on it. I also added ncloth on the skirt and binded the joints to the skirt. Then I try to let the leg to collide the skirt . The skirt collides fine, simulating cloth well. But I found the joint on the skirt didn't move at all, which leads that I can't get the skirt joint output data. So this problem confused me these days. Can you tell me some tips about it.Thank you ver much!
@@azagoruyko hi,I tried to train my network but the loss was very high.Just like what you said. I only used upper/lower legs's location to predict the joint's rotation and location.My model is torch.nn.Linear(D_in, H, bias=True), torch.nn.Tanh(), torch.nn.Linear(H, D_out, bias=True) D_in is 6 [left_leg_upper_x,left_leg_upper_ y, left_leg_upper_z, left_leg_lower_x,left_leg_lower_ y, left_leg_lower_z], D_out is 6 [skirt_joint_x, skirt_joint_y, skirt_joint_z, skirt_joint_pitch, skirt_joint_yaw, skirt_joint_raw] H = 50 and tanh is activation function. Did I foget some important operations?
This is really amazing! Can you tell me exactly how to do it? I really want to learn.
I have also watched the teaching of binding skirts you did before, and it is really very good.
Thank you very much.
Hi! Thanks a lot! Start with a super book called "Make Your Own Neural Network" by Tariq Rashid
@@azagoruyko Thank you very much.
epic!
good job. HUGE LIKE
Hi, It's really amazing. But I'm confused about the part of gathering data and trainning network. Does the data only contain information about skirt joint? Does neural network work this way:input the action of the character's bones and predict the joint information of the skirt?Can you tell me more information about these. Thank you very much.
Hi! Inputs are leg joints' X axis (X axis for upper/lower legs, 3*4 doubles), Outputs are skirt joints translations/rotations. What's confusing?
I use ncloth with skirt joints attached to it to train the network. I rotate legs by a lot of angles and calculate ncloth dynamics with my skirt joints.
So I have a lot of X axes for legs and skirt joints in each position. After I feed custom X axes and get desirable skirt joints' translation/rotation.
@@azagoruyko I Got it, impressive work. Thank you for your explanation. : )
@@azagoruyko Hi,After your explanation, I tried to implement it these days, But I have some troubles on gathering data.Can you help me? Here's my problem. I made a skirt by myself and added joints on it. I also added ncloth on the skirt and binded the joints to the skirt. Then I try to let the leg to collide the skirt . The skirt collides fine, simulating cloth well. But I found the joint on the skirt didn't move at all, which leads that I can't get the skirt joint output data. So this problem confused me these days. Can you tell me some tips about it.Thank you ver much!
@@howiewang8735 hmm. you have skirt joints attached to the ncloth surface? Just that shape. Use rivet script for this.
@@azagoruyko hi,I tried to train my network but the loss was very high.Just like what you said. I only used upper/lower legs's location to predict the joint's rotation and location.My model is
torch.nn.Linear(D_in, H, bias=True),
torch.nn.Tanh(),
torch.nn.Linear(H, D_out, bias=True)
D_in is 6 [left_leg_upper_x,left_leg_upper_ y, left_leg_upper_z, left_leg_lower_x,left_leg_lower_ y, left_leg_lower_z],
D_out is 6 [skirt_joint_x, skirt_joint_y, skirt_joint_z, skirt_joint_pitch, skirt_joint_yaw, skirt_joint_raw]
H = 50 and tanh is activation function.
Did I foget some important operations?
круто!!!