qinnzou / robust-lane-detection Goto Github PK

I did not find the code for multi-GPU training in your code. Could you please tell me how to run code with two or four GPUs at the same time

I am trying to run the code from scratch and trying to create my own model after training. Training successfully completes and i get a folder named 97.8457815.pth in the root directory. Upon extracting, this folder contains data folder, data.pkl and version folder. I want to ask which one is the trained model in these files. i considered data.pkl as trained model and save it in pretrained folder but upon executing test.py it unable to pick the trained model. whereas if i save your provided trained model i.e unetlstm, then it works fine. Moreover, in train.py file the trained model is saved at line 44 with the extension of .pth file but it is not saved like that in actual. Please guide me how can i get my trained model after completing the training of the model.

How could I make the test_index? Thanks.

Could you upload the complete text index?

How to increase image resolution? Current resolution (256,128) is too low.

Thanks,

你好，我发现您写的评估指标和使用混淆矩阵计算出来的评估结果差别比较大，我没搞明白是为啥？

please show the details "how to calculate json file from the predicted results of tusimple" at your convenience. thank you very much

When I executed the train.py and test.py . it leads error
ipykernel_launcher.py: error: unrecognized arguments: -f /home/jayamani/.local/share/jupyter/runtime/kernel-9bb45534-702f-4755-841c-299558dd0d1e.json
An exception has occurred, use %tb to see the full traceback.

Tools.py
name 'op_split_lstm' is not defined

I have downloaded your dataset from google drive.
What is the difference between train_index.txt and train_index_new.txt, val_index.txt and val_index_new.txt?

Which one did you use in your paper? I also find the files in your code. What is the difference between them? Which one should I use?

请问train_index.txt和val_index.txt可以给我发一下吗？应该怎么生成呢?

Hello, can you give me the dataset? There is no dataset in the download link.

I don't see a parameter in convlstm model below which allows me to specify how many frames to be used.
def init(self, input_size, input_dim, hidden_dim, kernel_size, num_layers,
batch_first=False, bias=True, return_all_layers=False):

Your current model is 5 and I am thinking to adjust this number to be higher or lower.

Thanks,

In which platform did u executed the code

I hope to add new train set but I don't know how to make label image with new load image.

Could you let me know how to draw the label image or which tools you used?

Hi, I saw that you have included in the README that you have pretrained model for ConvLSTM, UNet-ConvLSTM) as well as the comparable two(SegNet, UNet).
But I could only find pretrained model for UNet-ConvLSTM in the download link. Can you please share the other pretrained models?

I have looked in the links you provided from baidu and gdrive, it seems that "test_index.txt" file is missing. Could you advise me on this?

Why the SegNet_ConvLSTM and SegNet output F.log_softmax(up1, dim=1)?
The Cross Entropy loss function has already the softmax operation.

Hi, thanks a lot for publishing the code. It is a very interesting work. I am currently able to run the code with my own dataset. However, I notice that the latent space by LSTM is also an output for the training phase (test):

May I ask why the test is also considered when calculating the loss? Many thanks :)

Is there a typo on third row of the table?

Stride Sampled frames
3 1th 4th 7th 11th 13th
or
3 1th 4th 7th 10th 13th

/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/optim/lr_scheduler.py:100: UserWarning: Detected call of lr_scheduler.step() before optimizer.step(). In PyTorch 1.1.0 and later, you should call them in the opposite order: optimizer.step() before lr_scheduler.step(). Failure to do this will result in PyTorch skipping the first value of the learning rate schedule.See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
"https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate", UserWarning)
Traceback (most recent call last):
File "train.py", line 132, in
train(args, epoch, model, train_loader, device, optimizer, criterion)
File "train.py", line 18, in train
output = model(data)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/root/data/AIModel/Robust-Lane-Detection/LaneDetectionCode/model.py", line 53, in forward
x1 = self.inc(item)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/root/data/AIModel/Robust-Lane-Detection/LaneDetectionCode/utils.py", line 32, in forward
x = self.conv(x)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/root/data/AIModel/Robust-Lane-Detection/LaneDetectionCode/utils.py", line 22, in forward
x = self.conv(x)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/container.py", line 92, in forward
input = module(input)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 345, in forward
return self.conv2d_forward(input, self.weight)
File "/root/data/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 342, in conv2d_forward
self.padding, self.dilation, self.groups)
RuntimeError: CUDA out of memory. Tried to allocate 4.98 GiB (GPU 0; 15.90 GiB total capacity; 7.36 GiB already allocated; 1.07 GiB free; 6.77 GiB cached)

I have a 16GB GPU, but I keep getting this error.
I was wondering how you trained on two GPUs.
By the way, I am using Pytorch 1.3.1.
Thanks for your help!

Hi, thanks for the nice work. The groundtruth of the 13th frames are now jpgs, which are binary segmentation masks without any lane specific labels. I'm wondering if the groudthuth in the original tusimple json format is available. Or if the segmentation masks can be instance segmentation masks, that would also be very helpful. Thanks!

Is compare the single frame input with multi-frame input unfair?

What does these code mean as fellows in "test.py"? Will it make the evaluate results higher?
label_precision= cv2.dilate(lab, kernel) pred_recall = cv2.dilate(img, kernel)

12种不同场景下的图片都有标签，计算指标是只计算间隔五张的指标还是为每张图片生成预测结果再计算指标呢？可以为每张连续图片生成结果吗，还是只能五张五张生成结果？

Hi author, the evaluation metrics of your code after testing only contain Accuracy, Precision, Recall, F1-Measure, how to find out FP and FN based on these four values? You have presented the values of FP and FN in your paper. Can you share the code to find the FP and FN, thanks a lot!

您好，有一处问题想请教一下
lstm, _ = self.convlstm(data) test = lstm[0][-1,:,:,:,:]
第一行是定义了2层的lstm，一层用于提取特征，一层用于集成。
第二行的test =test = lstm[0][-1,:,:,:,:] 。 不知道是什么意思， 希望能解答一下， 谢谢

I'm using this model, but I want to use in 1280x720 images, how I can ? I'm trying to pic the binary result image and resizing to 1280x720, but the results are bad.