inhwanbae / enet-sad_pytorch Goto Github PK

Thank you for your open source, it is very profitable. I would like to ask the author what is the calculation formula of the BDD100K evaluation index pixel accuracy. The formula is not listed in the paper. Is it category pixel accuracy (acc = (TP) / TP + FP)?

bdd100k module is missing in dataset folder, can you provide the code for that.

thanks for your excellent work! I want to use the SAD module in my Wide-ResNet, but how could I separate out this module?

hi, thank you for the good job. When I train with my dataset, the sad loss is very small. Do you know why?

Traceback (most recent call last):
File "demo_video.py", line 184, in
main()
File "demo_video.py", line 120, in main
net.load_state_dict(save_dict['net'])
File "/home/zty/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 847, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for ENet_SAD:
size mismatch for fc.0.weight: copying a param with shape torch.Size([128, 4500]) from checkpoint, the shape in current model is torch.Size([128, 4400]).

Can somebody help me?Thanks in advance.

안녕하세요! 저는 생태 관련해서 연구하고 있는 석사과정 대학원생입니다.

다름이 아니라, 어쩌다보니 lane-detection에 대해 공부하고 있는 중인데, 관련 코드를 찾던중 선생님의 코드를 찾게되어 문의드립니다.

train.py의 경우에는 CULane에 대해서만 학습을 할 수 있는 것인가요?

그거랑 또 한가지 궁금한 점이, test를 하려면 반드시 CULane이나 tuSimple 데이터셋의 format과 동일하게 만들어야 test가 가능한지 궁금합니다.

왜냐하면 제가 학습/테스트를 하고자 하는 제 이미지 사이즈가 다소 차이가 좀 있어서, 그대로 테스트해도 되는지, 그리고 어노테이션을 할 수는 있는 것인지가 궁금합니다.

좋은 코드 만들어주셔서 감사합니다!

Hi,
I was digging into your code recently. Nice work!

Just had question about your CULane training dataset. It seems that you didnt use txt which contains axis for lanes in the given dataset.

Could your give a little hint why this would work?

Thanks!

There are 75 images that have 5 lanes but the architecture is to detect 5 categories (background + 4 lanes). I have removed those images from training and used the images (3193) that have less than 5 lanes. In the paper 3268 images are used for training. Can someone explain if those images were used or not and if they were used what was the approach taken?

In the Paper, it can reach 13.4ms on titan X, but why your code is 16ms on 2080ti, shouldn’t it be faster? Hope to answer it, thank you very much

Hi, I want to trian the net based on bdd100k, What should I pay attention to? For example, setting hyperparameters or Is there anything that needs to be changed in train.py. Looking forward to your reply

I check your Enet-SAD-model.py,and I found that the lane prediction branch in your code is different from the paper,can you tell me if there is any reason for this?Thanks in advance.

Hi, thanks for the great PyTorch implementation. I was wondering how long it might take for you to make the testing code public? Thanks!

우선 좋은 코드를 공유해주셔서 감사합니다.

혹시 BDD100K를 평가할때 사용한 코드도 공유 가능한가요?

감사합니다.

Is it normal that it takes me 30min to train an epoch on 1080ti?

and the test time is slow too.

Where is the pre training weight given，pls?

Hey, I'm using this repository for some research. I've trained the model (the only change I made to parameters was lowering batch size to 5 for my gpu), and I'm having trouble replicating the model's results. I am capping out at 85.8% accuracy. Do you have a pre-trained model available? Would really appreciate it :)

I'm meeting the following issue while using Tensorflow 2.0

AttributeError: module 'tensorboard.summary._tf.summary' has no attribute 'FileWriter'

您好，您可以提供一下训练的权重吗~
我这边自己训练的权重，测试出来很多图片的lines.txt都是空的，在多个机器上试了是相同的结果。可以提供一下您的权重，进行测试吗

Hey I'm very grateful for your work, and now I further need your codes for testing and evaluation, can you please provide it? Thanks a lot!

Hi,
Thanks for sharing this nice repo. Would you please provide the evaluation code for BDD100K dataset? Or give an advice where to find such code? This would be a great help. Thanks!