Hello,
Your paper is very interesting. Thank you for sharing your code. I have a question regarding Line 218 in image_editor_zecon.py file
https://github.com/YSerin/ZeCon/blob/51e679ef49d3875893eb956411abc802d9d8fc4c/optimization/image_editor_zecon.py#L218C1-L219C1
From what I understand, out['pred_xstart'] is \hat(x_0(x_t)). Since we have x0 available, the loss can be computed between out['pred_xstart'] and x0. Why have you taken a weighted combination in Line 218 to obtain x_t and y_t before computing the loss.
Also, could you please point me to the part of the paper that discusses the formulation of the 'fac' term in line 218.
regards,
hemanth
Hello,
Your paper is very interesting. Thank you for sharing your code. I have a question regarding Line 218 in image_editor_zecon.py file
https://github.com/YSerin/ZeCon/blob/51e679ef49d3875893eb956411abc802d9d8fc4c/optimization/image_editor_zecon.py#L218C1-L219C1
From what I understand, out['pred_xstart'] is \hat(x_0(x_t)). Since we have x0 available, the loss can be computed between out['pred_xstart'] and x0. Why have you taken a weighted combination in Line 218 to obtain x_t and y_t before computing the loss.
Also, could you please point me to the part of the paper that discusses the formulation of the 'fac' term in line 218.
regards,
hemanth