hazirbas / poselstm-pytorch Goto Github PK

Hi there,

Thanks for this repo.

Do you know any software or python code which stitches the point clouds together using the poses estimated by pose network?

Cheers,
Jacob

Hi,
I saw in the paper that your implementation is with Tensorflow.
Since the result of pythorch is not that promising, did you release the tensorflow implementation of your work?
Is there any difference between this implementation and the paper?
Thanks

It is a great implementation of the two camera pose estimation papers. Thank you for your work! It would be even better if you could offer the trained parameters of PoseNet and PostLSTM :-)

Hi. Thank you for your work. I've been trying to run the PoseNet model on a different dataset. The image size is 2880x2160 [HxW]. So I firest resize and save the images with the compute_mean_image.py and then use the argument --loadSize and --fineSize when running train.py to have the image square cropped to 512x512. But I get an error which I haven't been able to trace back.

But when I don't use the --fineSize argument (which I believe should be cropping to the default size of 224), the model runs fine.

CMD:
python util/compute_image_mean.py --dataroot /dataroot/ --height 720 --width 540 --save_images
python train.py --model posenet --dataroot /dataroot/ --name posenet/Hyundai/beta500 --beta 500 --gpu 0 --loadSize 540 --fineSize 512

ERROR:
posenet initializing the weights from pretrained_models/places-googlenet.pickle ---------- Networks initialized ------------- model [PoseNetModel] was created Traceback (most recent call last): File "train.py", line 38, in <module> model.optimize_parameters() File "/home/varora/PythonProjects/posenet/poselstm-pytorch/models/posenet_model.py", line 101, in optimize_parameters self.forward() File "/home/varora/PythonProjects/posenet/poselstm-pytorch/models/posenet_model.py", line 76, in forward self.pred_B = self.netG(self.input_A) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/varora/PythonProjects/posenet/poselstm-pytorch/models/networks.py", line 219, in forward return self.cls1_fc(output_4a) + self.cls2_fc(output_4d) + self.cls3_fc(output_5b) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/varora/PythonProjects/posenet/poselstm-pytorch/models/networks.py", line 106, in forward output = self.cls_fc_pose(output.view(output.size(0), -1)) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/modules/container.py", line 119, in forward input = module(input) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/modules/linear.py", line 94, in forward return F.linear(input, self.weight, self.bias) File "/home/varora/conda4.3.30/envs/posenet/lib/python3.8/site-packages/torch/nn/functional.py", line 1753, in linear return torch._C._nn.linear(input, weight, bias) RuntimeError: mat1 dim 1 must match mat2 dim 0

Any help is appreciated.
Thank you.

The link of pretrained model is unable opened，could you publish it again？Thank you very much.

Hi @hazirbas,

congrats for this implementation. Great job.

According to the architecture of the network:

I believe that the lstm_hidden_size should be 128 and not 256 as it currently is. What do you say? Do you want me to do a pull request to fix it, or there is some reason that I'm not seeing for the 256?

Hello, can you provide a model of possenet that you trained well with pytorch? Thank you

Hi, I am getting this weird pickle error, any ideas on how to fix it? I was able to download the pretrained weights and tried to run the python train.py file (with all the options as specified in the main page)

Traceback (most recent call last): File "C:\Users\joshu\OneDrive\Desktop\Lightpoint\pose_estimation\ImageTools\posenet-pytorch\train.py", line 33, in <module> Traceback (most recent call last): File "<string>", line 1, in <module> for i, data in enumerate(dataset): File "C:\Users\joshu\OneDrive\Desktop\Lightpoint\pose_estimation\ImageTools\posenet-pytorch\data\custom_dataset_data_loader.py", line 43, in __iter__ File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\spawn.py", line 116, in spawn_main for i, data in enumerate(self.dataloader): File "C:\Users\joshu\anaconda3\envs\arvis\lib\site-packages\torch\utils\data\dataloader.py", line 438, in __iter__ exitcode = _main(fd, parent_sentinel) return self._get_iterator() File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\spawn.py", line 126, in _main File "C:\Users\joshu\anaconda3\envs\arvis\lib\site-packages\torch\utils\data\dataloader.py", line 386, in _get_iterator return _MultiProcessingDataLoaderIter(self) self = reduction.pickle.load(from_parent) File "C:\Users\joshu\anaconda3\envs\arvis\lib\site-packages\torch\utils\data\dataloader.py", line 1039, in __init__ EOFError: Ran out of input w.start() File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\process.py", line 121, in start self._popen = self._Popen(self) File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\context.py", line 224, in _Popen return _default_context.get_context().Process._Popen(process_obj) File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\context.py", line 336, in _Popen return Popen(process_obj) File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\popen_spawn_win32.py", line 93, in __init__ reduction.dump(process_obj, to_child) File "C:\Users\joshu\anaconda3\envs\arvis\lib\multiprocessing\reduction.py", line 60, in dump ForkingPickler(file, protocol).dump(obj) AttributeError: Can't pickle local object 'get_posenet_transform.<locals>.<lambda>'

Hi,
how did you draw the class activation map (CAM) in the paper? I mean that CAM is often used for a classification problem and how does it work for a regression problem (i.e. when the output vector is composed of continuous values rather than probabilities)?

Thank you for your help! when I am training my data, I can not understand one of those params, beta refers. You have written in README.txt that different scenes use the different beta, but I do not know which number exactly fits my dataset. my dataset is a 9m*6m football grassland.
plus, I am not a native English speaker, there may be something I expressed is not very clear.QAQ