Chen Yang

I do not think it's an error, criterion_identity is a mse loss in pixel-level, it is hard to minimize the loss(because of unpair images).

maybe you can try this code: ```` # Normalization parameters for pre-trained PyTorch models mean = np.array([0.485, 0.456, 0.406]) std = np.array([0.229, 0.224, 0.225]) def denormalize(tensors): """ Denormalizes image tensors...

maybe you can try my test code: ``` import argparse import os import numpy as np import math import itertools import datetime import time import torchvision.transforms as transforms from torchvision.utils...

感谢您快速而准确的回复:) 我之前也跑过Seedformer,发现也是复现不出他论文的精度,自己训的话会差一些结果,当时问了他们也没搞明白为什么? 我觉得可能是不是我下载的PCN数据集有问题,所以都没复现出开源权重的精度,我再去检查一下本地的数据集. 我同意你的观点,Upsample Transformer视乎不是一个可以稳定训练的网络,至少在Seedformer开源出来的代码来看,他是魔改的SnowflakeNet,但是训的并不稳定. 有没有可能是默认的训练Lr=0.001太大了,我注意到pointr用的是0.0005/0.0001? --- 再次感谢您的回复, 祝您新年快乐~

Maybe you need to preprocess the dataset, I wrote a code to preprocess it and put it under ./data ```python import os import glob file_path="./hist2tscript" for img in glob.glob(file_path+"/*"): if...

BUT, I still lack "*_croods.tsv", if anybody has a solution? @OnkarMulay

@andrewbrown1022 我没有这个肿瘤标记数据,我现在是直接拿全部 spots 的基因表达训练

I have the same problem > FileNotFoundError: [Errno 2] No such file or directory: '/home/chenyang/ST-Net/stnet/utils/ensembl.tsv' could anyone help me? my email is [email protected]

Thanks @jiangdydy a lot, I download the emseml.tsv from https://www.genenames.org/cgi-bin/download/custom?col=gd_hgnc_id&col=gd_app_sym&col=gd_app_name&col=md_ensembl_id&status=Approved&status=Entry%20Withdrawn&hgnc_dbtag=on&order_by=gd_app_sym_sort&format=text&submit=submit you can download it

兄弟们,要不要加个微信,讨论怎么复现代码? 我的微信是: cyang23333