ACVNet
ACVNet copied to clipboard
gangweiX
[CVPR 2022] ACVNet: Attention Concatenation Volume for Accurate and Efficient Stereo Matching
Thanks for the great work. Could you also share the pretrained model on the KITTI dataset?
作者您好，原谅我使用汉语请教您，（我英文水平太菜了）。 1、对于SceneFlow数据集的评估，普遍都使用EPE（也就是MAE）作为评估标准，而且代码里也可以实现评估函数进行评估。 2、对于KITTI2012数据集，评价标准有Noc和Occ（All）的>2px, >3px, >4px, >5px以及Mean Error的错误率和错误像素数的评估，这些评估都是需要在自己代码里面实现它们的函数吗？还是需要提交到KITTI官网上生成评测结果呢？ 3、对于KITTI2015数据集，评价标准里有All（Occ）和Noc的D1-bg，D1-fg，D1-all的错误率评估，需要自己在代码里面实现评估函数吗，还是必须提交到KITTI官网上评测结果呢？ 3、发论文的话，KITTI12和15的实验数据必须来自kitti的官方网站吗？ 对于以上问题，目前还是比较迷惑的，kitti网站好像说是不能用于调试程序，每个人只能在规定时间内提交一次把，也不能申请多个账号吧。 所以对于这些评价标准的问题，还望作者大佬您能在百忙中抽出时间不吝赐教，万分感谢！！！~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~小白先行谢过！！！！！！！
关于数据集
作者您好，SceneFlow中包含了三个数据集FlyingThings3D、Driving、Monkaa，每个数据集又对应各自的RGB图片和disparity，请问训练的时候需要下哪个数据集呢？三个子集都要下载吗？以及下的是三个子集的RGB images (finalpass)吗？ FlyingThings3D | FlyingThings3D | | Driving | Driving | | Monkaa | Monkaa | -- | -- | -- | -- | -- | -- | --...
Hello, your project didn't include the evaluation code of different datasets? If I do a retrain, how to get the evaluation data of different datasets?
Hi, Thanks for the wonderful work. The code seems to have codes only for training the network. Do you have any plan to release any source code for demonstrations? I...
First of all, thank you very much for your very good work！！！ I tried to add an **inference()** function to **main.py** to check the inference effect. But the result is...
Hi, great work and thanks for providing your code. Do you plan on releasing the weights for the model finetuned on KITTI? Also, did you freeze the attention weights during...
thanks for your great work！ when i doing the first step,"python main.py --attention_weights_only True",there is always an error with loop output “creating new summary file loading model ./pretrained_model/pretrained_model_sceneflow.ckpt start at...
This network is excellent! I'd like to train on my own data, but my rectified images have positive and negative disparities. I see you use a max_disp, but is it...
