Xiao Pan

Which package does cmd ``fidelity`` rely on ?

2

Hi, ``fidelity`` is used in the training script for evaluating FID and KID. May I ask which specific package does this cmd reply on? Best,

iLIDS, VIPeR 数据集下载404 not Found

1

您好： 提供的iLIDS 和 VIPeR的下载链接失效了，能否提供百度云或其他网盘的下载链接？谢谢！ 祝好！

over half of the url are expired, could you please offer the download link for the dataset? thx

5

over half of the url are expired, could you please offer the download link for the dataset? we really need the dataset for academic research, thx!

The code for calculating chamfer distance.

1

Hi, Thank you for sharing the code. Could please also provide the code for calculating chamfer distance, as reported in the paper? Thx. Best,

在使用‘all’模式时，forward的前几行对batch_frame进行了去0，也就是说每个GPU上的图片数量应该是几个GPU中数量最少的那个，思路没有问题。但是我有两点疑问： 1.为什么不在前面的collate_fn中直接按照每个GPU输入帧数的最小值来shape，而是要先补0，然后到forward里去掉呢？ 2.为什么在下面这段代码中 if batch_frame is not None batch_frame = batch_frame[0].data.cpu().numpy().tolist() _ = len(batch_frame) for i in range(len(batch_frame)): if batch_frame[-(i + 1)] != 0: break else: _ -= 1 batch_frame...

阅读代码发现这个feature_num始终是等于1的，那么请问一下它的作用是什么呢？ 如果可以不是1的话，modle.fit()中的 feature, label_prob = self.encoder(*seq, batch_frame)肯定会报错把，因为在后面Gaitset只给了一个silho参数输入,但是*seq会提供多个参数。 希望您能解答我的疑惑，看了很久没有想明白。。谢谢！

i dont understand why during mutiGPUS trainning ,lr should multiply numgpus,thx!

How long does it take for training a video?

1

Hi, May I ask how long does it take to train a video? Best,

Hi, In ``3d_reconstruction.py``, you set the min and max for x, y, z manually. May I ask how to set these values under new scenes? Best,

YT-VOS 2018 or 2019 ?

1

Hi, May I ask which version you used for training? (2018 or 2019) And did you use the ```train_all_frames.zip``` or ```train.zip``` ? Thx.