| import numpy as np | |
| import torch | |
| def get_proj_mat(intrins, rots, trans): | |
| K = np.eye(4) | |
| K[:3, :3] = intrins | |
| R = np.eye(4) | |
| R[:3, :3] = rots.transpose(-1, -2) | |
| T = np.eye(4) | |
| T[:3, 3] = -trans | |
| RT = R @ T | |
| return K @ RT | |
| def perspective(cam_coords, proj_mat): | |
| pix_coords = proj_mat @ cam_coords | |
| valid_idx = pix_coords[2, :] > 0 | |
| pix_coords = pix_coords[:, valid_idx] | |
| pix_coords = pix_coords[:2, :] / (pix_coords[2, :] + 1e-7) | |
| pix_coords = pix_coords.transpose(1, 0) | |
| return pix_coords | |
| def label_onehot_decoding(onehot): | |
| return torch.argmax(onehot, axis=0) | |
| def label_onehot_encoding(label, num_classes=4): | |
| H, W = label.shape | |
| onehot = torch.zeros((num_classes, H, W)) | |
| onehot.scatter_(0, label[None].long(), 1) | |
| return onehot | |
| def gen_dx_bx(xbound, ybound, zbound): | |
| dx = torch.Tensor([row[2] for row in [xbound, ybound, zbound]]) | |
| bx = torch.Tensor([row[0] + row[2] / 2.0 for row in [xbound, ybound, zbound]]) | |
| nx = torch.LongTensor([(row[1] - row[0]) / row[2] for row in [xbound, ybound, zbound]]) | |
| return dx, bx, nx | |