jonathonluiten / premvos Goto Github PK

Hi,

When trying to run simple_run.sh script I am getting :

ModuleNotFoundError: No module named 'correlation_package._ext.corr._corr'

And then a bunch of Tracebak after:

'creating testnet...
using bbox guidance
network:
'

Here is atached the full log. It is quite lengthy !!

Thanks in advance
simple_run.log

Hi, could you tell me the difference between general_wight and specific_weights? thk a lot!

Dear,

You wrote

To use this, the original Lucid Data Dreaming code needs to be downloaded from here:

https://github.com/ankhoreva/LucidDataDreaming

And then it can be run with my script (run.py) to generate image augmentations per image sequence.

Could you tell me where is your run.py???

Thanks in advance

Hello,
How can I train this model especially the Proposals Generation?

Thanks for your nice code.
I'd like to train the ReID_net using my own dataset. But it seems like that the input of this net is not the original image format.
Could you provide some guidance of this part or maybe provide the code to generate the input datas?
Thank you.

I am running this repository with the following versions:

python 3.5.2
tensorflow 1.8
CUDA 7.5 (My GPU does not support more)
pytorch 0.2.0_3 (For cuda 7.5)

When I run the simple_run.sh it fives me the final values 0 0 0 0, apart from that it only gives warnings, and future warnings, and an index error.

The only changes I made to the code to been able to run ruin it without other errors were:

-change the library for the imread to matplotlib.pyplot, in this website tells that this is the workaround to the error scipy.misc not having that function.
-Ad the paths to PYTHONPATH to detect the functions inside the codes folder

The different parts of the simple run gives me the following results, the whole results are in the txt appended
results_PREMVOS.txt

:

##################### GENERATING COMBINED PROPOSALS
0.0001811981201171875
0.00019884109497070312

##################### GENERATING REFINED PROPOSALS
Several future warning similar to: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.

I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA

0 90 0 90
0 90 0 90 90 90
pre-prop-load
post-prop-load
0 90 0 90
0 90 0 90 90 90

creating testnet...
using bbox guidance
network:
testnet output : Using SEGMENTATION_LABELS_ORIGINAL_SIZE for calculating IoU
loading model from ../weights/PReMVOS_weights/refinement_net/specific_weights/refinement_specific_weights

Loading data from video

##################### GENERATING ReID PROPOSALS

Several future warnings similar to: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.

{
images darta (Not included to make shorter the issue)

"network": {
network data (Not included to make shorter the issue)
}

Traceback (most recent call last):
File "./ReID_net/main.py", line 41, in
tf.app.run(main)
File "/home/user/environment/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 126, in run
_sys.exit(main(argv))
File "./ReID_net/main.py", line 32, in main
engine = Engine(config)
File "/home/user/git/PReMVOS/code/ReID_net/Engine.py", line 86, in init
self.valid_data = load_dataset(config, "valid", self.session, self.coordinator)
File "/home///git/PReMVOS/code/ReID_net/datasets/Loader.py", line 88, in load_dataset
return DAVISForwardSimilarityDataset(config, subset, coordinator)
File "/home/user/git/PReMVOS/code/ReID_net/datasets/Similarity/DAVIS_Forward_Similarity.py", line 47, in init
super(DAVISForwardSimilarityDataset, self).init(config, subset, coord, annotations, n_train_ids=1)
File "/home/user/git/PReMVOS/code/ReID_net/datasets/Similarity/Similarity.py", line 58, in init
has_tags = "tag" in annotations[0]
IndexError: list index out of range

################# Merging

Several future warnings similar to: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.

creating testnet...
using bbox guidance
network:
testnet output : Using SEGMENTATION_LABELS_ORIGINAL_SIZE for calculating IoU
loading model from ../weights/PReMVOS_weights/refinement_net/specific_weights/refinement_specific_weights
../output/logs/ReID_net ReID_general_weights .log
number of parameters: 124,824,040
loading model from ../weights/PReMVOS_weights/ReID_net/general_weights/ReID_general_weights
0 0 0 0

there is no pytorch 0.2.0_3 binary with cuda 9.0: https://pytorch.org/get-started/previous-versions/.

Actually I have problem finding the perfect environment running this. My system is ubuntu 16.04 with cuda 8.0. I installed pytorch 0.2.0_3 and tensorflow 1.8.0 cpu version because it's not compatible with cuda 8.0, but got the error message "InvalidArgumentError (see above for traceback): Default MaxPoolingOp only supports NHWC on device type CPU" when I run sh simple_run.sh. Looks like the code has operations not supported by tensorflow cpu.

I also tried installing tensorflow 1.8.0 using docker. Since cuda version is 9.0 there, I could not install pytorch 0.2.0_3, and PWC-Net cannot be compiled with Pytorch 0.4

Hi, I want to fine-tune the method on a new video and I've learnt from your ACCV paper that there are three parts that were fine-tuned: Category Agnostic Mask R-CNN (proposal_net), Refinement Network and ReID Network.

However, I did not find any code in the proposal_net folder that performs the fine-tune operation and I was wondering if you could provide any help? Thanks!

I'm afraid this is a typo: https://github.com/JonathonLuiten/PReMVOS/blob/master/simple_run.sh#L40 should be PROP_NET_SPECIFIC_WEIGHTS=./weights/PReMVOS_weights/proposal_net/specific_weights/proposal_specific_weights
instead.

Hi, I'm running the docker image and after I run simple_run.sh, it outputs only selected folders and I can't find the scores. Can you please point me how to get the F and J scores? Thanks

Hi , I have read a recent paper:BoLTVOS:Box-LevelTrackingforVideoObjectSegmentation. This work uses the weight of pre-trained refine network to perform box2seg. So can you provide the pre-trained network? Thanks

+--------+--------+----------+---------+--------+----------+---------+
| Method | J_mean | J_recall | J_decay | F_mean | F_recall | F_decay |
+--------+--------+----------+---------+--------+----------+---------+
| final | 0.429 | 0.434 | 0.095 | 0.439 | 0.405 | 0.095 |
+--------+--------+----------+---------+--------+----------+---------+
Hi, Jonathon, after running the scripts in simple_run.sh, I indeed get the final results. Yet as shown in the table, the performances on sequences in Davis 2017 val.txt are poor, and far from the best results as reported in your papers, any tips for me? Thanks!

I am running this repository as it is in github with the following configuration:

python 3.6
tensorflow 1.8
pytorch 0.2.0_3 (pip install https://download.pytorch.org/whl/cu80/torch-0.2.0.post3-cp36-cp36m-manylinux1_x86_64.whl)
CUDA 9.0

which are the same as the described requirements, but whem triying to run an error appears:

Traceback (most recent call last):
  File "./code/optical_flow_net-PWC-Net/script_pwc_multi.py", line 9, in <module>
    import models
  File "/home/usuario/code/PReMVOS/code/optical_flow_net-PWC-Net/models/__init__.py", line 1, in <module>
    from .PWCNet import *
  File "/home/usuario/code/PReMVOS/code/optical_flow_net-PWC-Net/models/PWCNet.py", line 13, in <module>
    from correlation_package.modules.corr import Correlation 
  File "/home/usuario/code/PReMVOS/code/optical_flow_net-PWC-Net/external_packages/correlation-pytorch-master/correlation-pytorch/correlation_package/modules/corr.py", line 2, in <module>
    from ..functions.corr import correlation, correlation1d
  File "/home/usuario/code/PReMVOS/code/optical_flow_net-PWC-Net/external_packages/correlation-pytorch-master/correlation-pytorch/correlation_package/functions/corr.py", line 3, in <module>
    from .._ext import corr
  File "/home/usuario/code/PReMVOS/code/optical_flow_net-PWC-Net/external_packages/correlation-pytorch-master/correlation-pytorch/correlation_package/_ext/corr/__init__.py", line 3, in <module>
    from ._corr import lib as _lib, ffi as _ffi
ModuleNotFoundError: No module named 'correlation_package._ext.corr._corr'

I know there is a past issue closed with this same problem but I just tried to re-install torch with the link provided and with the one in the pytorch web. I also tried running it from cuda 8, cuda 9 and cuda 10

I want to finetune refinement_net using my datasets , but i am a bit confused about the code and konw nothing about how to load my own data as I dont kown the structrue of the data you used.
What kind of data needed for training? Bounding boxes of RGB images and groundtruth masks?
Could you provide a brief instruction to tell me how to train the net from the general weights?

I got the following errors when I run simple_run.sh

Traceback (most recent call last):
  File "./code/proposal_net/train.py", line 24, in <module>
    from coco import COCODetection
  File "/mnt/lustre/xiehaozhe/Development/PReMVOS/code/proposal_net/coco.py", line 11, in <module>
    from tensorpack.utils.rect import FloatBox  
ModuleNotFoundError: No module named 'tensorpack.utils.rect'