Please refer to https://www.zhangsongyang.com/ for more details.
tonysy / deep-feature-flow-segmentation Goto Github PK
View Code? Open in Web Editor NEWDeep Feature Flow for Video Semantic Segmentation
License: MIT License
Deep Feature Flow for Video Semantic Segmentation
License: MIT License
Please refer to https://www.zhangsongyang.com/ for more details.
您好,我在训练的时候,一直提示index out of range,就是predict label那一块,我发现在在您的代码里有很多使用了res_image_name = seg_pathes[1][:-len('_gtFine_labelTrainIds.png')]
的代码,这个_gtFine_labelTrainIds.png
我在cityscape中没有看到,只看到了_gtFine_labelIds.png
,想问一下您代码中的名字是根据感兴趣的类别自己重新生成的训练图像吗(label id一一映射过去,不感兴趣的类别都设置为255)?谢谢!
Can someone help me? I have a problem, I have changed the data set from label to labelTrainId,and start training, After starting the training, the following problem arose:
Epoch[0] Batch [290] Speed: 1.49 samples/sec Train-FCNLogLoss=1.237416,
Epoch[0] Batch [300] Speed: 1.51 samples/sec Train-FCNLogLoss=1.209671,
Epoch[0] Batch [310] Speed: 1.51 samples/sec Train-FCNLogLoss=1.188998,
Epoch[0] Batch [320] Speed: 1.51 samples/sec Train-FCNLogLoss=1.172847,
Epoch[0] Batch [330] Speed: 1.51 samples/sec Train-FCNLogLoss=1.152629,
Epoch[0] Batch [340] Speed: 1.52 samples/sec Train-FCNLogLoss=1.139912,
Epoch[0] Batch [350] Speed: 1.51 samples/sec Train-FCNLogLoss=1.121668,
libpng error: Read Error
Exception in thread Thread-9:
Traceback (most recent call last):
File "D:\Software\Anaconda3\envs\FGF\lib\threading.py", line 801, in __bootstrap_inner
self.run()
File "D:\Software\Anaconda3\envs\FGF\lib\threading.py", line 754, in run
self.__target(*self.__args, **self.__kwargs)
File "./experiments/deeplab....\deeplab..\lib\utils\PrefetchingIter.py", line 60, in prefetch_func
self.next_batch[i] = self.iters[i].next()
File "./experiments/deeplab....\deeplab\core\loader.py", line 188, in next
self.get_batch_parallel()
File "./experiments/deeplab....\deeplab\core\loader.py", line 237, in get_batch_parallel
rst = [multiprocess_result.get() for multiprocess_result in multiprocess_results]
File "D:\Software\Anaconda3\envs\FGF\lib\multiprocessing\pool.py", line 572, in get
raise self._value
ValueError: zero-size array to reduction operation minimum which has no identity
Can someone help me?thanks.
[11:10:10] src/base.cc:51: Upgrade advisory: this mxnet has been built against cuda library version 9000, which is older than the oldest version tested by CI (10000). Set MXNET_CUDA_LIB_CHECKING=0 to quiet this warning.
learning rate from lr_scheduler
has been overwritten by learning_rate
in optimizer.
Traceback (most recent call last):
File "./experiments/deeplab/deeplab_train_test.py", line 23, in
train.main()
File "./experiments/deeplab/../../deeplab/train.py", line 209, in main
config.TRAIN.begin_epoch, config.TRAIN.end_epoch, config.TRAIN.lr, config.TRAIN.lr_step)
File "./experiments/deeplab/../../deeplab/train.py", line 203, in train_net
arg_params=arg_params, aux_params=aux_params, begin_epoch=begin_epoch, num_epoch=end_epoch)
File "./experiments/deeplab/../../deeplab/core/module.py", line 976, in fit
self.update_metric(eval_metric, data_batch.label)
File "./experiments/deeplab/../../deeplab/core/module.py", line 1068, in update_metric
self._curr_module.update_metric(eval_metric, labels)
File "./experiments/deeplab/../../deeplab/core/module.py", line 665, in update_metric
self._exec_group.update_metric(eval_metric, labels)
File "./experiments/deeplab/../../deeplab/core/DataParallelExecutorGroup.py", line 490, in update_metric
eval_metric.update(labels, texec.outputs)
File "/usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/metric.py", line 363, in update
metric.update(labels, preds)
File "./experiments/deeplab/../../deeplab/core/metric.py", line 28, in update
pred = pred.asnumpy().reshape((pred.shape[0], pred.shape[1], -1)).transpose((0, 2, 1))
File "/usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/ndarray/ndarray.py", line 2535, in asnumpy
ctypes.c_size_t(data.size)))
File "/usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/base.py", line 255, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: [11:10:13] src/operator/fusion/fused_op.cu:604: Check failed: compileResult == NVRTC_SUCCESS (5 vs. 0) : NVRTC Compilation failed. Please set environment variable MXNET_USE_FUSION to 0.
nvrtc: error: invalid value for --gpu-architecture (-arch)
Stack trace:
[bt] (0) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x6b41eb) [0x7f78887981eb]
[bt] (1) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x44fb5a6) [0x7f788c5df5a6]
[bt] (2) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x4505d87) [0x7f788c5e9d87]
[bt] (3) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x37d69b9) [0x7f788b8ba9b9]
[bt] (4) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x37e33d5) [0x7f788b8c73d5]
[bt] (5) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x37bf6d1) [0x7f788b8a36d1]
[bt] (6) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x37c2c10) [0x7f788b8a6c10]
[bt] (7) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x37c2ea6) [0x7f788b8a6ea6]
[bt] (8) /usr/local/anaconda3/envs/py27/lib/python2.7/site-packages/mxnet/libmxnet.so(+0x37bde84) [0x7f788b8a1e84]
I installed mxnet-cu90 and meet this problem,Can anyone help me?Thank you .
Hi, what should i put in the folder ./data/cityscapes_video/gtFine_sequence_own? The gtFine_trainvaltest ground truth? I have downloaded the leftImg8bit_sequence but there is no ground truth for the sequence. In the paper of Deep-Feature-Flow the authors say they have used one ground truth image for each sequence. Did you do any kind of pre-processing to the original ground truth in order to adapt it to the sequewnce dataset?
Thank you
Hi, have you reproduced the DFF work on semantic segmentation task? I would be very thankful if you can release your code.
Excuse me, could you show me some hints about how to test on my own dataset since you had set the segmentation label for all your training and testing images? thx
thank you for providing this implementation of segmentation!
from the rfcn implementation of the original paper, the key frame interval is set under the config. how to set the interval in your implementation?
That's so kind of you if you release that..the registration of the cityscape website requires few days...
What's the result of your deeplab baseline on cityscape? Thank you!
Thanks for the code updates you have made. Glad to see deep featurte flow for semantic segmentation.
Kindly let us know do you have inference codes written for video sequence? Kindly let us know.
张博士,你好:
看到你在readme中写到,如果想要训练DFF,使用Cityscapes Video Data。我在Cityscapes中只找到了一个demovideo,而且好像也没有标签。不知道想要训练DFF做视频语义分割的话,该如何设置数据集呢?是否还是使用gtFine_trainvaltest这个文件,将train,val,test三个图片文件放入data/cityscapes中进行训练呢?还望您不吝赐教,万分感谢。
when choosing current image and other neighbor image to train dff, why set the probability of choosing current frame and other neighbor frame the same? As written in image.py(224)
Hi @tonysy ,
I was trying to run DFF using
python ./experiments/deeplab_dff/deeplab_dff_train.py --cfg ./experiments/deeplab_dff/cfgs/deeplab_resnet_v1_101_cityscapes_segmentation_video.yaml
For data preparation, I have downloaded gtFine_trainvaltest.zip (241MB) and leftImg8bit_trainvaltest.zip (11GB) and the folder has structure like
data/cityscapes_video/gtFine
data/cityscapes_video/leftImg8bit
However, when I try DFF it outputs error for missing data as following:
======== Step:1,Starting to load_gt_segdb =========
======== Total Number of images: 2975======
======== Starting to get gt_segdb ========
======== Starting to create gt_segdb ======
100%|███████████████████████████████████████| 2975/2975 [02:16<00:00, 21.73it/s]
========= Wrote gt segdb to ./data/cache/cityscape_video_leftImg8bit_train_gt_segdb.pkl
=====Start appending flipped images to segdb(./lib/imdb)=====
100%|█████████████████████████████████████| 2975/2975 [00:00<00:00, 6716.08it/s]
{'data': (1L, 3L, 768L, 1024L),
'data_ref': (1L, 3L, 768L, 1024L),
'eq_flag': (1L,),
'label': (1L, 1L, 768L, 1024L)}
./model/pretrained_model/resnet_v1_101
lr 0.0005 lr_epoch_diff [40.336, 60.504] lr_iters [59999, 89999]
ref image path error, image not exists! ./data/cityscapes_video/leftImg8bit/train/tubingen/tubingen_000111_000016_leftImg8bit.png
ref image path error, image not exists! ./data/cityscapes_video/leftImg8bit/train/weimar/weimar_000100_000016_leftImg8bit.png
ref image path error, image not exists! ./data/cityscapes_video/leftImg8bit/train/weimar/weimar_000095_000016_leftImg8bit.png
ref image path error, image not exists! ./data/cityscapes_video/leftImg8bit/train/erfurt/erfurt_000102_000016_leftImg8bit.png
These image files are not stored in the zip file I downloaded from Cityscape website. I want to ask what kind of data you used to run DFF when you mentioned to download cityscapes video data and put it into data/cityscapes_video? Thank you very much.
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.