| |
| |
| |
| |
| |
| |
|
|
| """Functions for ramping hyperparameters up or down |
| |
| Each function takes the current training step or epoch, and the |
| ramp length in the same format, and returns a multiplier between |
| 0 and 1. |
| """ |
|
|
|
|
| import numpy as np |
| import torch |
|
|
| def norm_image(image, ep = 1e-8): |
| range = torch.max(image)-torch.min(image) |
| if range == 0: |
| norm_image = image |
| else: |
| norm_image = (image-torch.min(image))/(range+ep) |
| return norm_image |
|
|
| def get_imgs(p_first, p_second, p_new, volume_batch, label_batch, sample_index): |
| ins_width = 2 |
| num_colums = 7 |
| B,C,H,W,D = p_first.size() |
| snapshot_img = torch.zeros(size = (D, 3, num_colums * H + num_colums * ins_width, W + ins_width), dtype = torch.float32) |
|
|
| for icol in range(1, num_colums+1): |
| snapshot_img[:,:, icol*(H+ins_width)-ins_width:icol*(H+ins_width),:] = 1 |
| snapshot_img[:,:, :,W:W+ins_width] = 1 |
|
|
| seg_out_1 = p_first[sample_index,0].permute(2,0,1).cpu().data |
| seg_out_2 = p_second[sample_index,0].permute(2,0,1).cpu().data |
| seg_out_3 = p_new[sample_index,0].permute(2,0,1).cpu().data |
|
|
| target = label_batch[sample_index].permute(2,0,1).cpu().data |
|
|
| train_img_1 = volume_batch[sample_index,0].permute(2,0,1).cpu().data |
| train_img_2 = volume_batch[sample_index,1].permute(2,0,1).cpu().data |
| |
| for i_rgb in range(3): |
| snapshot_img[:, i_rgb,:H,:W] = norm_image(train_img_1) |
| snapshot_img[:, i_rgb, H+ ins_width:2*H+ ins_width,:W] = norm_image(train_img_2) |
| snapshot_img[:, i_rgb, 2*H+ 2*ins_width:3*H+ 2*ins_width,:W] = norm_image(train_img_2-train_img_1) |
| snapshot_img[:, i_rgb, 3*H+ 3*ins_width:4*H+ 3*ins_width,:W] = seg_out_1 |
| snapshot_img[:, i_rgb, 4*H+ 4*ins_width:5*H+ 4*ins_width,:W] = seg_out_2 |
| snapshot_img[:, i_rgb, 5*H+ 5*ins_width:6*H+ 5*ins_width,:W] = seg_out_3 |
| snapshot_img[:, i_rgb, 6*H+ 6*ins_width:7*H+ 6*ins_width,:W] = target |
|
|
| return snapshot_img |
|
|
| def sigmoid_rampup(current, rampup_length): |
| """Exponential rampup from https://arxiv.org/abs/1610.02242""" |
| if rampup_length == 0: |
| return 1.0 |
| else: |
| current = np.clip(current, 0.0, rampup_length) |
| phase = 1.0 - current / rampup_length |
| return float(np.exp(-5.0 * phase * phase)) |
|
|
|
|
| def linear_rampup(current, rampup_length): |
| """Linear rampup""" |
| assert current >= 0 and rampup_length >= 0 |
| if current >= rampup_length: |
| return 1.0 |
| else: |
| return current / rampup_length |
|
|
|
|
| def cosine_rampdown(current, rampdown_length): |
| """Cosine rampdown from https://arxiv.org/abs/1608.03983""" |
| assert 0 <= current <= rampdown_length |
| return float(.5 * (np.cos(np.pi * current / rampdown_length) + 1)) |
|
|
