CoderXuans

Results 24 issues of CoderXuans

@deepcs233 Hello, I want to ask about the fine-tuning that I plan to implement on the cloud server. For the final deployment on my local machine, what level of GPU...

@deepcs233 Hello, how can I remove the suddenly appearing pedestrians? In which folder should I modify to achieve this?

@deepcs233 I want to replace the ego vehicle with an unmanned boat during evaluation. Should I make changes from both aspects: the code and the simulation?

My colleague provided me with an FBX file for the exterior. Can I replace only the exterior of the existing vehicle in CARLA without changing the original vehicle dynamics model?

@deepcs233 你好,我试着把预测出来的点打印出来,发现它们距离很近也不太稳定,这种情况用PID是最好的解决办法,所以不使用模型预测控制吗?

@deepcs233 你好,我在一台GPU上运行train.sh之后, 1、报File "train_pretrain.py", line 821, in __call__ prob_det = torch.sigmoid(output[:, :, 0] * (1 - 2 * target[:, :, 0])) IndexError: too many indices for tensor of dimension 2这个问题...

@deepcs233 你好,我看到视觉编码器的train.sh脚本中有 --pretrained \ 一行,我理解的是它能够用你提供训练好的模型权重,在新数据集上进行微调,这样理解对吗?

@deepcs233 你好,我想请问一下if self.has_qformer: print('Loading Q-Former') self.Qformer, self.query_tokens = self.init_Qformer( 4, self.visual_encoder.num_features ) self.Qformer.resize_token_embeddings(len(self.llm_tokenizer)) self.Qformer.cls = None 中的4如果改为别的数字的话,需要自己重新执行Instruction finetuning的训练吗? 我理解的Instruction finetuning是让LLM适应驾驶任务。

@deepcs233 你好,我自己的电脑是一张4090。 1、我能用它进行视觉编码器预训练以及指令微调吗? 2、我最开始尝试过租云服务器,但是云服务器的存储空间只有50G放数据集,我只想在town05中运行评估,这种情况下只用town05的数据集可以吗?我知道使用的数据集越多得到的训练效果越好。

@deepcs233 你好,请问可以用NuScenes数据集对视觉编码器进行评估么?