依赖项更新

.github/workflows/ci.yaml

@@ -25,5 +25,5 @@ jobs:
           path: .cache
           restore-keys: |
             mkdocs-material-
-      - run: pip install mkdocs-material[imaging]==9.5.47 mkdocs-git-revision-date-localized-plugin==1.3.0 mkdocs-git-authors-plugin==0.9.2 mkdocs-static-i18n==1.2.3
+      - run: pip install mkdocs-material[imaging]==9.5.48 mkdocs-git-revision-date-localized-plugin==1.3.0 mkdocs-git-authors-plugin==0.9.2 mkdocs-static-i18n==1.2.3
       - run: mkdocs gh-deploy --force

README.md

@@ -144,7 +144,7 @@ python tools\export_pt.py --cfg C:\workspace\github\monolite\experiment\monolite
 
 # Confirm致谢
 
-我们衷心感谢所有为这个神经网络开源项目做出贡献的个人和组织。特别感谢以下贡献者：
+我们衷心感谢所有为这个神经网络开源项目做出贡献的个人和组织。特别感谢：
 
 | type       | name             | url                                                 | title                                                                    |
 | ---------- | ---------------- | --------------------------------------------------- | ------------------------------------------------------------------------ |

lib/datasets/kitti_utils.py

@@ -576,6 +576,17 @@ def roty(t):
     s = np.sin(t)
     return np.array([[c, 0, s], [0, 1, 0], [-s, 0, c]])
 
+def rotx(t):
+    """Rotation about the x-axis."""
+    c = np.cos(t)
+    s = np.sin(t)
+    return np.array([[1, 0, 0], [0, c, -s], [0, s, c]])
+
+def rotz(t):
+    """Rotation about the z-axis."""
+    c = np.cos(t)
+    s = np.sin(t)
+    return np.array([[c, -s, 0], [s, c, 0], [0, 0, 1]])
 
 if __name__ == "__main__":
     from lib.datasets.kitti import KITTI

lib/utils/metrics.py

@@ -3,7 +3,8 @@
 import torch.nn.functional as F
 import math
 import numpy as np
-from typing import Optional,Union
+from typing import Optional, Union
+
 
 def bbox_iou(box1, box2, xywh=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7):
     """
@@ -121,31 +122,36 @@ def xyxy2xywh(x):
 def filter_boxes(boxes, image_width, image_height):
     """
     Filter out boxes that are outside the image boundaries.
-    
+
     Parameters:
     - boxes (Tensor): Tensor of shape (n, 4) where each row is [x_min, y_min, x_max, y_max].
     - image_width (int): The width of the image.
     - image_height (int): The height of the image.
-    
+
     Returns:
     - Tensor: Filtered boxes that are within the image boundaries.
     """
     # 确保boxes是PyTorch张量
     boxes = torch.as_tensor(boxes, dtype=torch.float32)
-    
+
     # 检查边界框的坐标是否在图像范围内
     # x_min 和 y_min 应该大于等于0，x_max 和 y_max 应该小于等于图像的宽高
-    condition = (boxes[:, 0] >= 0) & (boxes[:, 1] >= 0) & \
-                (boxes[:, 2] <= image_width) & (boxes[:, 3] <= image_height)
-
+    condition = (
+        (boxes[:, 0] >= 0)
+        & (boxes[:, 1] >= 0)
+        & (boxes[:, 2] <= image_width)
+        & (boxes[:, 3] <= image_height)
+    )
+
     # 同时x_max 应该大于 x_min, y_max 应该大于 y_min
     condition = condition & (boxes[:, 2] > boxes[:, 0]) & (boxes[:, 3] > boxes[:, 1])
-    
+
     # 根据条件筛选出合法的边界框
     valid_boxes = boxes[condition]
-    
+
     return valid_boxes
 
+
 def crop_3d_points(
     points: np.ndarray,
     x_range: tuple[float, float],
@@ -169,16 +175,39 @@ def crop_3d_points(
     return cropped_points
 
 
+def inside_test(points, cube3d):
+    """
+    cube3d  =  numpy array of the shape (8,3) with coordinates in the clockwise order. first the bottom plane is considered then the top one.
+    points = array of points with shape (N, 3).
+
+    Returns the indices of the points array which are outside the cube3d
+    """
+    b1, b2, b3, b4, t1, t2, t3, t4 = cube3d
+
+    # 计算三个单位法向量dir1、dir2、dir3
+    dir1 = t1 - b1
+    size1 = np.linalg.norm(dir1)
+    dir1 = dir1 / size1
+
+    dir2 = b2 - b1
+    size2 = np.linalg.norm(dir2)
+    dir2 = dir2 / size2
+
+    dir3 = b4 - b1
+    size3 = np.linalg.norm(dir3)
+    dir3 = dir3 / size3
+
+    # 计算中心点cube3d_center
+    cube3d_center = (b1 + t3) / 2.0
+
+    # 点到中心点cube3d_center的向量
+    dir_vec = points - cube3d_center
+
+    # dir_vec到3个单位法向量的投影距离，再乘2
+    # 小于box三条边的长度，则符合要求
+    res1 = np.where((np.absolute(dir_vec @ dir1) * 2) <= size1)[0]
+    res2 = np.where((np.absolute(dir_vec @ dir2) * 2) <= size2)[0]
+    res3 = np.where((np.absolute(dir_vec @ dir3) * 2) <= size3)[0]
 
-def rotation_matrix_ry(theta:Union[int,float,np.ndarray])->np.ndarray:
-    # 将角度转换为弧度
-    theta_rad = np.radians(theta)
-
-    # 创建旋转矩阵
-    rotation_matrix = np.array([
-        [np.cos(theta_rad), 0, np.sin(theta_rad)],
-        [0, 1, 0],
-        [-np.sin(theta_rad), 0, np.cos(theta_rad)]
-    ])
-
-    return rotation_matrix
+    # 符合的结果取交集
+    return np.array(list(set(res1) & set(res2) & set(res3)))

requirements.txt

@@ -5,17 +5,17 @@ torchinfo==1.8.0
 opencv-python==4.10.0.84
 pillow==10.4.0
 psutil==6.1.0
-matplotlib==3.9.3
+matplotlib==3.10.0
 pyecharts==2.0.7
 open3d==0.18.0 # https://www.open3d.org/docs/latest/getting_started.html
 #grad-cam==1.5.4
-albumentations==1.4.21
+albumentations==1.4.22
 
 # log tools (must be installed)
-loguru==0.7.2
+loguru==0.7.3
 rich==13.9.4
-swanlab==0.3.27
-#tqdm==4.66.5
+swanlab==0.4.0
+#tqdm==4.67.1
 
 # export tools (optional)
 onnx==1.17.0
@@ -33,14 +33,14 @@ monkeytype==23.3.0
 # optimizer tools (optional)
 # https://mmengine.readthedocs.io/zh-cn/latest/common_usage/better_optimizers.html
 prodigyopt==1.0
-lion-pytorch==0.2.2
+lion-pytorch==0.2.3
 Sophia-Optimizer==0.2.5
 
 # hyperparameter (optional)
 optuna==4.1.0
 
 # doc tools (optional)
-mkdocs-material[imaging]==9.5.47
+mkdocs-material[imaging]==9.5.48
 mkdocs-git-revision-date-localized-plugin==1.3.0
 mkdocs-git-authors-plugin==0.9.2
 mkdocs-static-i18n==1.2.3
@@ -49,5 +49,8 @@ mkdocs-static-i18n==1.2.3
 pytest==8.3.4
 pytest-cov==6.0.0
 
+# performance anylysis tools (optional)
+# scalene==1.5.49 # Not a eazy way to use
+
 # check if torch.complie is supported on windows
 # https://github.com/woct0rdho/triton-windows