请问您的工作是先生成一个像素级的bev特征，然后根据像素级bev得到极点bev特征吗

[fyctime052]

Thank you so much for your interest in PolarPoint-BEV. For your question, the answer is no, the polar point BEV map is not generated based on the pixel-level BEV map. For your reference, please check the #PolarPointBEV/modules.py, in which shows the generation process of polar point BEV map. Thank you.

[May525116]

Based on your answer to the previous question, I checked the #PolarPointBEV/modules.py file and you defined three classes to perform feature transformation on the input multi-view feature map and decode the feature map into a class probability distribution. But I didn't find anything about "conversion method of polar coordinate BEV map". Can you help provide more accurate information? Where or how did you convert the BEV map into polar coordinate representation? Very much looking forward to your reply!

[May525116]

Specifically, I would like to know more details about the module (BEV decoder), and I hope you can help point me to the specific location of this module in the code. Very much looking forward to your reply!

[1146065545]

Specifically, I would like to know more details about the module (BEV decoder), and I hope you can help point me to the specific location of this module in the code. Very much looking forward to your reply!

有关于您提到的decoder，module.py里有一句提到self.decoder = Graph(num_class)，不知道是否是您问题的答案。

[May525116]

非常感谢您的关注和回复，这个Graph类的确是用作了decoder，我的理解是，这个decoder只是用于将特征映射到了一个概率分布，比如他分了三个类，每一个特征对应的类别的概率是多少。我疑惑的点在于如何得到了极坐标表示，就是他流程图里面展示的，经过decoder之后，这些特征点（原文说是三类语义点）如何转换成了一个极坐标的表示方法。从Graph类里面的 nn.AdaptiveAvgPool2d((16, 27)),# (16, 27) for normal; (16, 15) for sparse; (16, 21) for light; (16, 33) for thick; (16, 41) for dense 这部分，也仅仅能看出是一个矩形的表达。

[fyctime052]

Thank you for your comment. The 2D matrix (for example, (16, 27) for normal configuration) generated in the Graph class is the polar point BEV map. There is no explicit process of converting to polar coordinates. In the training process, the ground truth of the polar point BEV map is also expressed in a 2D matrix, which is the same as the output of Graph class. I hope this answers your question.

[fyctime052]

BTW, a 2D matrix does not necessarily represent the Cartesian coordinate system. Take the 2D matrix of (16, 27) for example, the 16 represents the number of layer of polar point BEV map, 27 represents the number of points in each layer.

[May525116]

Thank you very much for your answers and replies!！I already understand the problem！！！ Because I am a beginner, there may be other problems later. If you have time, I hope the author will give you some advice.