zhangjy328
zhangjy328
Does this model have any requirements for the size of the input images? Can it be used for large-sized images?
Hello, author. I want to use my own dataset to perform some style transfer tasks, such as converting land scenes into an underwater style. However, I only want to transfer...
Hello, experts. I have tried multiple models to denoise my images, but there is no effect at all before and after denoising. For example, when testing the DNCNN model, I...
你好，我想在我自己的数据集上训练模型，该如何实现？对数据集有什么格式要求？如果我只有图片，没有深度信息可以无监督学习吗？
Hello, author. I have checked the pseudo ground truth in the NYU dataset and the model that generates the pseudo ground truth. The result of the pseudo ground truth is...
所以air数据集的作用就是训练GAN网络，这个用什么数据对结果没影响，直接用提供的就好？water数据集使用自己的数据就可以吗？它是通过把air中的图片生成成water的效果，再反向恢复water图片吗？
作者你好， 我想请教一个关于特征编码的问题，我看到dataset里`cld`是完全随机选取的`cld = dpt_xyz.reshape(-1, 3)[choose, :]`，后续knn的最近邻也是依据随机的`cld`进行计算的，在特征编码的地方`p2r_emb = self.nearest_interpolation(p2r_emb, p2r_ds_nei_idx)`，`p2r_ds_nei_idx`也是完全随机选取的点，虽然存在映射关系，但是随机选取很有可能出现分布不均的情况，为什么可以直接view成`rgb_emb0`的形状，跟`rgb_emb0`进行cat呢？ 1.`cld`完全随机，计算最近邻的时候，某些点最近邻的点完全有可能跟自身无关，如果它周围的点都没被随机到的话 2.如果`cld`里选取的点都比较聚集，那么`p2r_emb` 的特征就是局部的，直接view位置和信息并不对应，为什么能跟`rgb_emb0`进行cat？比如640×480图像，极端一点，`cld`随机的全是前120行的点，`p2r_emb` 对应的76800就是640×120的信息，直接view成320×240不是全都错位了吗？`rgb_emb0`下采样到320×240对应的也是全图信息，直接cat也不对吧，能帮我解释一下吗？ Dear Author, I would like to ask a question regarding feature encoding. I noticed that in the...
Dear Author, I would like to ask: Is this model only capable of training on data with a fixed number and type of items? If each image contains a different...
When I am training, there is always a "nan" prompt, but I can't find where this prompt is printed and don't know what the error is. The same happens during...
I want to ask, can I train the model if I only have one GPU? How to set it up? By the way, can this model run on Windows?