zhechen / plard Goto Github PK
View Code? Open in Web Editor NEWProgressive LiDAR Adaptation for Road Detection
License: MIT License
Progressive LiDAR Adaptation for Road Detection
License: MIT License
Hi,
After running the test with the pre-trained model, the output results are stored in /output/results.
By using the evaluation code from devkit_road download from KITTI website, the following is the result:
Starting evaluation ...
Available categories are: ['um_lane', 'um_road', 'umm_road', 'uu_road']
Execute evaluation for category um_lane ...
Searching for submitted files with prefix: um_lane_
Computing evaluation scores...
MaxF: 86.93
AvgPrec: 90.74
PRE_wp: 86.91
REC_wp: 86.95
FPR_wp: 1.17
FNR_wp: 13.05
Finished evaluating category: um_lane
Execute evaluation for category um_road ...
Searching for submitted files with prefix: um_road_
Computing evaluation scores...
MaxF: 89.62
AvgPrec: 93.21
PRE_wp: 89.32
REC_wp: 89.93
FPR_wp: 2.12
FNR_wp: 10.07
Finished evaluating category: um_road
Execute evaluation for category umm_road ...
Searching for submitted files with prefix: umm_road_
Computing evaluation scores...
MaxF: 83.00
AvgPrec: 89.42
PRE_wp: 77.65
REC_wp: 89.14
FPR_wp: 8.03
FNR_wp: 10.86
Finished evaluating category: umm_road
Execute evaluation for category uu_road ...
Searching for submitted files with prefix: uu_road_
Computing evaluation scores...
MaxF: 81.50
AvgPrec: 87.35
PRE_wp: 79.03
REC_wp: 84.13
FPR_wp: 3.52
FNR_wp: 15.87
Finished evaluating category: uu_road
Successfully finished evaluation for 4 categories: ['um_lane', 'um_road', 'umm_road', 'uu_road']
Apparently, the result is far from what you posted in your paper.
Would you please help me figure out why?
Btw, a lot of bugs are found in the evaluation code, and I have to correct them in order to run.
Hi,
Thank you very much for your code. Could I know which model is used as the pre-trained model in your training?
The PSP model on CitySpaces or the resnet101 pre-trained on ImageNet.
Many thanks
Hi,
Now, I can replicate the result of the test using GPUs.
However, when I train the model with the following command:
python train.py --arch plard --dataset kitti_road --no-visdom --pretrained ./plard_kitti_road.pth
I get the error:
Params: l_rate 0.000050, l_rate_decay: 0.10, l_rate_step: 1, batch_size: 4, mom: 0.99, wd: 0.000500
Using custom loss
/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/functional.py:2416: UserWarning: nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.
warnings.warn("nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.")
/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/_reduction.py:13: UserWarning: reduction='elementwise_mean' is deprecated, please use reduction='mean' instead.
warnings.warn("reduction='elementwise_mean' is deprecated, please use reduction='mean' instead.")
Traceback (most recent call last):
File "train.py", line 185, in
train(args, logger)
File "train.py", line 101, in train
loss = model([images, lidars, labels])
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/modules/module.py", line 532, in call
result = self.forward(*input, **kwargs)
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 152, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 162, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/parallel/parallel_apply.py", line 85, in parallel_apply
output.reraise()
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/_utils.py", line 394, in reraise
raise self.exc_type(msg)
RuntimeError: Caught RuntimeError in replica 0 on device 0.
Original Traceback (most recent call last):
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/parallel/parallel_apply.py", line 60, in _worker
output = module(*input, **kwargs)
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/nn/modules/module.py", line 532, in call
result = self.forward(*input, **kwargs)
File "/home/u4067/Documents/PLARD/PLARD_intact/ptsemseg/models/plard.py", line 358, in forward
loss.backward()
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/tensor.py", line 195, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/home/u4067/anaconda3/envs/py_plard/lib/python3.7/site-packages/torch/autograd/init.py", line 99, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: CUDA out of memory. Tried to allocate 120.00 MiB (GPU 0; 7.79 GiB total capacity; 6.37 GiB already allocated; 116.00 MiB free; 6.53 GiB reserved in total by PyTorch)
Could you please help with this error?
My env:
ubuntu: 16.4
python: 3.7.0
pytorch: 1.4.0
torchfile: 0.1.0
torchversion: 0.2.0
GPU: Geforce RTX 2070
cuda: 10.0
And changing the batch size doesn't help.
I tried to get the summary of the model. For That we needed input size.
On investigation it was found out that the shape of inputs are different for RGB and LiDAR Model.
RGB Model has -> (1,3,384,1280)
Lidar Mode has -> (1,1,384,1280)
Pleas confirm what should be the input shape for the model so that Model summary can be shown.
Thanks
Which lidar sensor you have used very good code thank you
I am just trying to use my own calibration data using your ADI code in matlab. but it does not work properly. May I ask how did you re-generate calibration format of Kitti? how did you calibrate your camera and lidar? My calib info are coming from autoware lidar-camera calibration and transforming it to kitti format has not been successful.
Hello, where can i find the relevant information of NF2CNN network.
Original code without changing anything.
But LOSS stacks at around 0.45 after 20 epoches.
Could you help me?
Have tried LR from 1e-4 to 1e-5
Got very bad results using Test.py and a trained model with 0.45 loss.
感谢你的开源,最近在试着训练你的网络,发现loss并不下降,从1左右迅速下降到0.5之后(不到一个epoch),就不再下降,而且训练出来的权重,测试结果完全没法看,铁别差。环境和你的一样,请问这大概是什么原因呢?谢谢!
Hi,
When I run the code, I got:
RuntimeError: CuDNN error: CUDNN_STATUS_SUCCESS.
Would you please help find out what the problem is?
My env:
ubuntu: 16.4
python: 3.7.0
pytorch: 0.4.1
torchfile: 0.1.0
torchversion: 0.2.0
GPU: Geforce RTX 2070
cuda: 10.0
Thank you for your open source!Now i need use your trained model to test my image,but your trained model in here can't download.So please tell me what can i do.Thanks again.
Thank you for your work,when i train the model,I can't find the definition of the fllowing functions in train.py line 44:model.load_state_dict_without_classification(checkpoint['model_state']). and line 98: model.module.freeze_bn().Is there any missing document?Looking forward to your reply.
Hi,
Thank you very much for your code.
I use the pspnet101_cityscapes pre-trained model to train kitti, but I cannot achieve the results of the paper. Can you provide suggestions?
dataset: kitti training 289 sheets
model: PLARD (no modification)
batch_size: 8
Learning Rate: 1e-4 -> 1e-6 (SGD)
epoch: 80
Many thanks
thanks a lot for this work.
Could you also share the code of Data space adaptation step?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.