oppo-us-research / PlaneMVS

Official PyTorch implementation of PlaneMVS ( https://arxiv.org/abs/2203.12082 ) , a 3D plane reconstruction framework leveraging multi-view geometry and slanted plane hypotheses.
MIT License
10 stars 0 forks source link

运行python test_net.py遇到的一系列问题请教 #3

Open Okay0805 opened 2 hours ago

Okay0805 commented 2 hours ago

首先,测试数据取了三对图:(先开始只放了一对图,后来又加了两对) O9)1@N3MUL{}8}_O 6$@%5H 预训练模型先开始是准备用model_final_exp03.pth的,但是报错RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory:(model_final_exp02.pth也出现相同的情况) image 后来用的model_final_exp04.pth,遇到RuntimeError:(model_final_exp01.pth也出现相同的情况) @0GIJ@2O}B2`X{IAO)7}V35 然后去修改planstereo.yaml中的LOSS_TERM_NUM,将9改为10之后没有报RuntimeError了,不知道这样修改对不对 M1D(MJIS2ZFPP3 0F443YRA

Okay0805 commented 2 hours ago

接下来是报错需要deltas_split_stereo_cleanedsegmentation ![LWIM4@834 _ 60Q0S41%LP](https://github.com/user-attachments/assets/0b4dd861-4b00-4ac6-bdb0-691561f5038f) 然后我想这是测试应该是不需要提供分割图的,于是在scannet_stereo.py中修改了self.is_test_split直接为True: V`DH{6%B`Q1 _HU`3N%YKG 然后就遇到了一直没找到解决方法的RuntimeError: Unknown layout: CMM_UCVB$J_7Z)(RWNIJ4 T 去planestereo.py中去查看输入的shape: image images['ref_img'] = torch.Size([3, 480, 640])和ref_img_tensors shape = torch.Size([1, 3, 480, 640]), torch.float32, device= cuda:0是正确的,不知道下一步该从何解决,请指教,谢谢!