PointCLIP

PointCLIP copied to clipboard

在评估完模型性能后，似乎没有获得每个测试样本的预测类别，而是只获得了accuracy的值。想知道如何输出每个预测值？ @ZiyuGuo99

LinLin1031

I wanted to ask if we can get bounding boxes for the classified objects in the point cloud using this approach.

ayaan-mphasis

Thanks for the amazing work! I modified the zero-shot classification code for ScanObjectNN, but only got 10.5% (15.38% in the paper) accuracy for the OBJ_ONLY variation. Is there anything missing...

kaimingkuang

您好，我在运行train.py文件的时候，在跳转到trainer文件的TrainerX(SimpleTrainer)，会遇到这个问题： for self.batch_idx, batch in enumerate(self.train_loader_x): data_time.update(time.time() - end) loss_summary = self.forward_backward(batch) batch_time.update(time.time() - end) losses.update(loss_summary) 其中： def forward_backward(self, batch): raise NotImplementedError 是这样定义的，所以我一运行就报这个错误，我的基础比较差，不知道能不能麻烦您解释一下，帮我看看这个问题

WSDW-wh

How to adapt PointCLIP for part-segmentation task

1

Hi, @ZrrSkywalker , Thanks for releasing the package and also the guide on ZSL for 3D object classification. I'm concerned about whether it's possible to adapt the PointCLIP method for...

amiltonwong

How to ensemble pointclip and other point model?

1

Hello author.I am very interested in your research.Could you tell me how to ensemble pointclip and other point model such as pointnet? Could you give me some tips?I'm looking forward...

IzhiSu

Hello！使用原码进行zero-shot分类测试，使用vit-b-32模型，modelnet40测试集复现top1 acc为16%左右，scanobjenn测试集为7%左右，与原文精度有一定差距，请问是否需要调节一些超参数设置呢？

vikyzeng

Hi, thanks for the interesting work! I noticed that, for the zeroshot classification, only the modelnet40 dataset part of code was released, is there any chance the scanobjectnn part of...

Tycho-Xue

点云的feature与clip中text feature的问题

1

想了解一下有考虑过用pointnet之类的学到的point cloud global feature加上全连接层做finetune然后与clip中的text feature比较这样的尝试吗，还是说因为clip中image encoder和text encoder学到的特征是对齐的，所以直接考虑了2d depth maps projection的思路，如果有考虑过前者的话，是效果不好吗？

jiatianzhi

Apply for the pretrained weights on other few-shot learning

1

Thanks for your wonderful work, it is very interesting and helpful to my research. Therefore, I want to apply the other few-shot learning pre-trained weights not only 16-shot as follows:...

wangye8899