| |
| import os, cv2 |
| import torch |
| import torch.nn.functional as F |
| import torch.distributed as dist |
| import math |
| import numpy as np |
| import PIL.Image as Image |
| import matplotlib.pyplot as plt |
|
|
| from torch.utils.data import Sampler |
| from torchvision import transforms |
|
|
| |
| def tensor2disp(tensor, vmax=0.18, percentile=None, viewind=0): |
| cm = plt.get_cmap('magma') |
| tnp = tensor[viewind, 0, :, :].detach().cpu().numpy() |
| if percentile is not None: |
| if np.sum(tnp > 0) > 100: |
| vmax = np.percentile(tnp[tnp > 0], 95) |
| else: |
| vmax = 1.0 |
| tnp = tnp / vmax |
| tnp = (cm(tnp) * 255).astype(np.uint8) |
| return Image.fromarray(tnp[:, :, 0:3]) |
|
|
| def tensor2grad(gradtensor, percentile=95, pos_bar=0, neg_bar=0, viewind=0): |
| cm = plt.get_cmap('bwr') |
| gradnumpy = gradtensor.detach().cpu().numpy()[viewind, 0, :, :] |
|
|
| selector_pos = gradnumpy > 0 |
| if np.sum(selector_pos) > 1: |
| if pos_bar <= 0: |
| pos_bar = np.percentile(gradnumpy[selector_pos], percentile) |
| gradnumpy[selector_pos] = gradnumpy[selector_pos] / pos_bar / 2 |
|
|
| selector_neg = gradnumpy < 0 |
| if np.sum(selector_neg) > 1: |
| if neg_bar >= 0: |
| neg_bar = -np.percentile(-gradnumpy[selector_neg], percentile) |
| gradnumpy[selector_neg] = -gradnumpy[selector_neg] / neg_bar / 2 |
|
|
| disp_grad_numpy = gradnumpy + 0.5 |
| colorMap = cm(disp_grad_numpy)[:, :, 0:3] |
| return Image.fromarray((colorMap * 255).astype(np.uint8)) |
|
|
| def tensor2rgb(tensor, viewind=0): |
| tnp = tensor.detach().cpu().permute([0, 2, 3, 1]).contiguous()[viewind, :, :, :].numpy() |
| if np.max(tnp) <= 2: |
| tnp = tnp * 255 |
| tnp = np.clip(tnp, a_min=0, a_max=255).astype(np.uint8) |
| return Image.fromarray(tnp) |
|
|
