i found that the Volatile GPU-Util is zero in most of time, so i want to know whether the data loading took a long time, but in pytorch ,which provide the "workers" parameter to set multithreading to read data,but it still slow ,i dont know the reason and hopeful to get the answer.

Sorry to interrupt again...I have done the training part and the checkpoint file is generated, and I want to do the inference task, so I write a demo using the .pth file and some frames cropped from a video in order to predict its class. But one error occurs:

Traceback (most recent call last):
  File "new_demo.py", line 105, in <module>
    main()
  File "new_demo.py", line 84, in main
    model.load_state_dict(checkpoint['state_dict'])
  File "/home//anaconda2/lib/python2.7/site-packages/torch/nn/modules/module.py", line 522, in load_state_dict
    .format(name))
KeyError: 'unexpected key "module.base_model.conv1_7x7_s2.weight" in state_dict

The code loading ckpt file is as follows:

checkpoint = torch.load(args.ckpt_file)
print checkpoint['state_dict'], type(checkpoint['state_dict'])
model.load_state_dict(checkpoint['state_dict'])

I have no idea what is wrong...

though the num_class is set at the begin of the main.py as follow ,but it seems not being used in the program, could you give me some suggestion ？
thank you~
-------------------main.py-------------------------
if args.dataset == 'ucf101':
num_class = 101
elif args.dataset == 'hmdb51':
num_class = 51
elif args.dataset == 'kinetics':
num_class = 400
elif args.dataset == 'something':
num_class = 174

@zhang-can 您好，非常感谢你的工作，我安装readme文档中的方法进行实验，训练和测试数据分别产生了三个不同的文件，假设是train_list1.txt,train_list2.txt,train_list3.txt和val_list1.txt,val_list2.txt,val_list3.txt,
请问在执行python main.py ucf101 RGB <ucf101_rgb_train_list> <ucf101_rgb_val_list>应该如何将这三个文件传递给训练程序呢？谢谢！

Hi, I have downloaded the ECOLite pre-trained models following the instructions given here https://github.com/mzolfaghari/ECO-pytorch.
I am not sure what is the difference between the two pre-trained ECOLite models: eco_lite_rgb_16F_kinetics_v3.pth.tar and ECO_Lite_rgb_model_Kinetics.pth.tar
Thanks!

Author, thank you very much for your work.
I have a question, I want to get the classification accuracy of the 101 class, but after the test my results have been like this, what do you think might be the problem?

[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Accuracy 0.99%

Looking forward to your reply

Hi, Can
Thanks so much for your contribution for this pytorch version.
I have tested on something dataset, I found the performance is around 3-4 worse than the caffe version. I obtain 38.6 accuracy using pytorch code ECO-lite with 16 segements whereas the caffe achieved 42.4 accuracy.

I have some questions about the difference between this pytorch version and original caffe version.

1. The mean is [104 117 123] in caffe whereas the pytorch version uses [104 117 128].
2. The caffe model resizes the image to [320 240] followed by cropping whereas the pytorch version resize the image to [256 256] followed by cropping.

Do you think these differences will result into performance difference?
Meanwhile, I wonder the pretrained model on Kinetics you provided is obtained by training your pytorch model or just convert from the pretrained caffe model?

Looking forward to your reply.
Best,
Tan

When using single input, model always produces wrong output

Hello,

I am Youngkyoon Jang, who is a postdoc in the University of Bristol. Thanks for sharing the PyTorch implementation of ECO network for testing new dataset, EPIC-KITCHEN. But I am facing the problem when testing the model to calculate the accuracy.

I am currently using PyTorch implementation for ECO network. And I noticed that the model always predicts the first index with the highest score when using a single input instead of multiple mini-batch samples in testing. Did you know this problem? Is there a correct way to get a consistent output?

When I put a different number of batch sample in a mini-batch, the model predicts a different score for the same sample depending on the number of mini-batch.

I look forward to your reply.

Best,
Young

currently training and test is ok, but if I change the args.num_segments from 4 to 8, an error will occur:

RuntimeError: size mismatch, m1: [24 x 1024], m2: [512 x 8] at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:249

where should I modify the code to solve this problem?

@zhang-can thanks!

Thank you for you code, it is really helpful to me.
recently I try to update the pretrain model from you new url,
'https://s3.us-east-2.amazonaws.com/zhangcan/kin_ECO_epoch-12-c89e9dc0.pth.tar'

but it hard to download from china, could you give us another option?

Thank you~

Hi, thank you for releasing pytorch version of ECO.

Have you run ECO-pytorch on the validation set of something-something-v1 dataset?

I wonder if it have reached the performance on the paper, 46.4%, which is obtained from an ensemble of networks with {16, 20, 24, 32} number of frames.

Thank you.

When I tried to run the command " sh run_demo_ECO_Full.sh local ", the "RuntimeError: CUDA error(2): out of memory" showed up.

Hi, I find that before input the tensor into the network, you don't use torch.autograd.Variable to convert a tensor to a Variable. How does this work?

I met a problem named as "FileNotFoundError: [Errno 2] No such file or directory: 'dataset\UCF-101\WalkingWithDog\img_00001.jpg' when I trained the model using "run_ECOLite_finetune_UCF101.sh".It couldn't continue to run with "train(train_loader, model, criterion, optimizer, epoch)" using debug.I could't find the reason resulting in this. can you help me ? Thanks!

Hi @zhang-can , thank you for your work! I can't access to Google Docs, so could you please upload the latest pretrained model eco_lite_rgb_16F_kinetics_v3.pth.tar on BaiduYun?

I did some expermients based on ECO,but encountered some problem about the benchmark datasets.My downloaded something -something dataset couldn't use with an unknown erro,so can anyone share the dataset?And the kinetics dataset is so bigger,I can't do the experiment on it(machine limiation).So how represent the scientific contrast with ECO and my modifying ?

How to use log.txt to draw loss and prec@1 picture?

Folks,

Can we train UCF Crime Dataset, I wanted to train Crime videos for a research. I see this model uses UCF 101.

Can you please let me know.

Thanks
Guru

Hi,

While running ECO-lite 4F I couldn't reach the reported results of top-1 85%. Instead got around 68%. Is there anything extra that needs to be done?

To my knowledge the only thing I'm doing differently is using a newer version of PyTorch.

Thanks

I would like to see the results of the ECO trained model. Please share with me. Thank you very much.
想看一下训练好的模型的结果，请大神分享一下，非常感谢。

sooorry to bother you again :p
I want to test the online recognition. but the inference erro like this: **RuntimeError: cuda runtime error (2) : out of memory at /pytorch/aten/src/THC/generic/THCStorage.cu:58**
only i set the batchsize to 2 and then it can work. but I find the eval will success when the batchsize is 15. below is the code:


I test it on UCF101. thanks in advande

Hello,

thank you for making this code available. you say that any contribution is welcome. I'd like to be help complete the code. could you add a TODO, so I know what tasks to start working on.

Closing this as I see that the code is being developed further in a different repo.

Thanks again for releasing this code

As reported in the paper of ECO, pretrained 2D BNInception and 3D resnet-18 models on Kinetics dataset are not enough to get a good result, training the ECO model for another 10 epochs on Kinectics would promise a better result. However, I am really not able to train the model on Kinectics(GPU and memory limitation). Since you mentioned that you are training the models on Kinetics, would you please share the trained model? I am going to use the trained weights to initialize the model and train it on something something dataset. I can report the testing result and share it in this repository.

When I tried to run the command " sh run_demo_ECO_Lite.sh local", the "RuntimeError: module must have its parameters and buffers on device cuda:1 (device_ids[0]) but found one of them on device: cpu" showed up.

Traceback (most recent call last):
File "main.py", line 601, in
main()
File "main.py", line 211, in main
train(train_loader, model, criterion, optimizer, epoch)
File "main.py", line 449, in train
output = model(input_var)
File "/data3/yjx/venv_eco/lib/python3.5/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/data3/yjx/venv_eco/lib/python3.5/site-packages/torch/nn/parallel/data_parallel.py", line 146, in forward
"them on device: {}".format(self.src_device_obj, t.device))
RuntimeError: module must have its parameters and buffers on device cuda:1 (device_ids[0]) but found one of them on device: cpu

AttributeError: module 'PIL.Image' has no attribute 'HAMMING'

scripts/run_ECOLite_finetune_UCF101.sh: 65: scripts/run_ECOLite_finetune_UCF101.sh: Syntax error: end of file unexpected (expecting "then")

thanks for your attenation

i has some ideas to write paper like eco,
but i donot know how to create the ECO.yaml from origin caffe model?

when I train ,I met a error called"Type error:'BatchNorm3d' object is not iterable" . I thought a lot about it ,but no outcome.

Hello, thank you for your contribution in the pytorch-ECO code in GitHub. However, I found some problems in the validate function of main.py.

It looks like 254 lines in main.py：

compute output

output = model(input_var)
loss = criterion(output, target_var)

It seems that the ECO network trained the validation data set. I hope you take the time to check it out during your busy schedule.

solved

good work!
When I did try this repo by using the dataset, something-something, there was a odd bug. Although I have all the data of the something under the running folder, including the
/20bn-datasets/20bn-something-something-v1/76110/00006.jpg' ,
the code will stop owing to FileNotFoundError: [Errno 2] No such file or directory: '../20bn-datasets/20bn-something-something-v1/76110/img_00006.jpg'.
Interestingly, this time is /76110/img-00006.jpg, while other time is /36265/img_00006.jpg. In Summary, it is changable!!
Thanks!
Looking forward to any answers!!
@zhang-can

first of all. thanks your code. i notice that you upload the ucf script sh . and I download the pretrained models by wget and the baidu disk. all failed. it show me that :

what's the result on testing set of UCF101 with 16 segments each video?

have a question that how to generate the *.yaml ,is handwriting? or generate this through other file.
thx : )

想看看这个ECO测试视频动作的效果，但是没有demo，作者能给一个sample吗？
If I want to test a video using trained models, how?

when I downloaded the something-something-v1 dataset , I surprisedly found that I couldn't get these files which consist of video and images . In contrary,I got some unkown type of files.DO has someone encountered the problem?

@zhang-can The code and dataset were the same as before, but I got the following error. However they could work well about a half mouth ago.
the pytorch's version is 0.3.1
raceback (most recent call last):
File "main.py", line 341, in
main()
File "main.py", line 141, in main
train(train_loader, model, criterion, optimizer, epoch)
File "main.py", line 192, in train
prec1, prec5 = accuracy(output.data, target, topk=(1,5))
File "main.py", line 329, in accuracy
_, pred = output.topk(maxk, 1, True, True)
RuntimeError: invalid argument 5: k not in range for dimension at /pytorch/torch/lib/THC/generic/THCTensorTopK.cu:21
/pytorch/torch/lib/THCUNN/ClassNLLCriterion.cu:101: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [0,0,0] Assertion t >= 0 && t < n_classes failed.
/pytorch/torch/lib/THCUNN/ClassNLLCriterion.cu:101: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [1,0,0] Assertion t >= 0 && t < n_classes failed.
/pytorch/torch/lib/THCUNN/ClassNLLCriterion.cu:101: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [2,0,0] Assertion t >= 0 && t < n_classes failed.
/pytorch/torch/lib/THCUNN/ClassNLLCriterion.cu:101: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [3,0,0] Assertion t >= 0 && t < n_classes failed.
THCudaCheck FAIL file=/pytorch/torch/lib/THC/generic/THCStorage.c line=184 error=59 : device-side assert triggered
terminate called after throwing an instance of 'std::runtime_error'
what(): cuda runtime error (59) : device-side assert triggered at /pytorch/torch/lib/THC/generic/THCStorage.c:184
Aborted (core dumped)

Thanks for your contribution!
I run the code successfully on Tesla M40, and it takes about 11 hours which is much longer than 3 hours mentioned in the origin paper. Is this normal?

你好，我想知道你的label序号是怎么设定的，因为我在预训练的模型上测试图片后并不清楚结果属于哪一类，谢谢！

I noticed that there are three train/test splits in UCF101 dataset, does it mean that we have three training and testing evaluation benchmarks?

您好，很感谢你提供的代码，现在有一个问题，我按照你的参数去训练，预训练模型用的是eco_lite_rgb_16F_kinetics_v3.pth.tar，但是loss一直不降低 维持在4.61左右。

你好，num_segments=4,numsegments=16画出来的网络结构是一样的，这是为啥呢?

When I set --consensus_type avg, I got a bug as following:
@zhang-can
Initializing TSN with base model: ECO.
TSN Configurations:
input_modality: RGB
num_segments: 4
new_length: 1
consensus_module: avg
dropout_ratio: 0.5

['fc_final.weight', 'fc_final.bias']
/data/yuyongbo/action_recognition/ECO-pytorch/tf_model_zoo/ECO/pytorch_load.py:60: UserWarning: nn.init.normal is now deprecated in favor of nn.init.normal_.
normal(new_state_dict[k], 0, std)
/data/yuyongbo/action_recognition/ECO-pytorch/tf_model_zoo/ECO/pytorch_load.py:62: UserWarning: nn.init.constant is now deprecated in favor of nn.init.constant_.
constant(new_state_dict[k], 0)
/data/yuyongbo/action_recognition/ECO-pytorch/models.py:91: UserWarning: nn.init.normal is now deprecated in favor of nn.init.normal_.
normal(self.new_fc.weight, 0, std)
/data/yuyongbo/action_recognition/ECO-pytorch/models.py:92: UserWarning: nn.init.constant is now deprecated in favor of nn.init.constant_.
constant(self.new_fc.bias, 0)
/home/yuyongbo/anaconda3/envs/py27-torch4/lib/python2.7/site-packages/torchvision-0.2.1-py2.7.egg/torchvision/transforms/transforms.py:188: UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.
group: first_conv_weight has 1 params, lr_mult: 1, decay_mult: 1
group: first_conv_bias has 1 params, lr_mult: 2, decay_mult: 0
group: normal_weight has 31 params, lr_mult: 1, decay_mult: 1
group: normal_bias has 19 params, lr_mult: 2, decay_mult: 0
group: BN scale/shift has 2 params, lr_mult: 1, decay_mult: 0
Freezing BatchNorm2D except the first one.
THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THC/THCCachingHostAllocator.cpp line=257 error=59 : device-side assert triggered
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [0,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [1,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [2,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [3,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [4,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [5,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [6,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [7,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [8,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [9,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [10,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [11,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [12,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [13,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [14,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [15,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [16,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [17,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [18,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [19,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [20,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [21,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [22,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [23,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [24,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [25,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [26,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [27,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [28,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [29,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [30,0,0] Assertion t >= 0 && t < n_classes failed.
/opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THCUNN/ClassNLLCriterion.cu:105: void cunn_ClassNLLCriterion_updateOutput_kernel(Dtype *, Dtype *, Dtype *, long *, Dtype *, int, int, int, int, long) [with Dtype = float, Acctype = float]: block: [0,0,0], thread: [31,0,0] Assertion t >= 0 && t < n_classes failed.
Exception NameError: "global name 'FileNotFoundError' is not defined" in <bound method _DataLoaderIter.del of <torch.utils.data.dataloader._DataLoaderIter object at 0x7fa0f247f3d0>> ignored
Traceback (most recent call last):
File "main.py", line 338, in
main()
File "main.py", line 137, in main
train(train_loader, model, criterion, optimizer, epoch)
File "main.py", line 188, in train
prec1, prec5 = accuracy(output.data, target, topk=(1,5))
File "main.py", line 325, in accuracy
_, pred = output.topk(maxk, 1, True, True)
RuntimeError: invalid argument 5: k not in range for dimension at /opt/conda/conda-bld/pytorch_1524577177097/work/aten/src/THC/generic/THCTensorTopK.cu:21
terminate called without an active exception
Aborted (core dumped)

y = self.con1x1(x) # 2D tensor
y = y.view((-1, 96, 16) + y.size()[2:]) #2D tensor changed to 3D tensor

error:

My environment is Ubuntu16.04 python2.7, torch.version '0.4.0' ;torchvision.version '0.2.1' . I have 8-gups with 1080Ti.
this repo will work will with one-GPU , however it will break down with multi-gups after run 1 epoch. The error is following:
Freezing BatchNorm2D except the first one.
main.py:251: UserWarning: volatile was removed and now has no effect. Use with torch.no_grad(): instead.
input_var = torch.autograd.Variable(input, volatile=True)
main.py:252: UserWarning: volatile was removed and now has no effect. Use with torch.no_grad(): instead.
target_var = torch.autograd.Variable(target, volatile=True)
main.py:261: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number
losses.update(loss.data[0], input.size(0))
main.py:262: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number
top1.update(prec1[0], input.size(0))
main.py:263: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number
top5.update(prec5[0], input.size(0))
Test: [0/46] Time 6.590 (6.590) Loss 4.9549 (4.9549) Prec@1 1.562 (1.562) Prec@5 9.375 (9.375)
THCudaCheck FAIL file=/pytorch/aten/src/THC/generic/THCStorage.cu line=58 error=2 : out of memory
Traceback (most recent call last):
File "main.py", line 338, in
main()
File "main.py", line 141, in main
prec1 = validate(val_loader, model, criterion, (epoch + 1) * len(train_loader))
File "main.py", line 255, in validate
output = model(input_var)
File "/home/yuyongbo/object_detection/venv-torch-latest/local/lib/python2.7/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/home/yuyongbo/object_detection/venv-torch-latest/local/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 114, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "/home/yuyongbo/object_detection/venv-torch-latest/local/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 124, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "/home/yuyongbo/object_detection/venv-torch-latest/local/lib/python2.7/site-packages/torch/nn/parallel/parallel_apply.py", line 65, in parallel_apply
raise output
RuntimeError: cuda runtime error (2) : out of memory at /pytorch/aten/src/THC/generic/THCStorage.cu:58

Thanks for you works on ECO, I just download your code and the DataSet ucf1101,
When I follow the instruction,I face such err:
could you help me check this error?
thanks~

================================ err ===================================

guo@guo-ubu:~/eng/ECO-pytorch-master$ python gen_dataset_lists.py ucf101 ~/eng/dataset/UCF-101/

Processing dataset ucf101:

• Parse frames under folder /home/guo/eng/dataset/UCF-101/

• 0/101 videos parsed

• Frame folder analysis done

• Writing list files for training/testing
  Traceback (most recent call last):
  File "gen_dataset_lists.py", line 100, in
  lists = build_split_list(split_tp, f_info, i, shuffle)
  File "/home/guo/eng/ECO-pytorch-master/pyActionRecog/benchmark_db.py", line 58, in build_split_list
  train_rgb_list, train_flow_list = build_set_list(split[0])
  File "/home/guo/eng/ECO-pytorch-master/pyActionRecog/benchmark_db.py", line 48, in build_set_list
  frame_dir = frame_info[0][item[0]]
  KeyError: 'v_ApplyEyeMakeup_g08_c01'
  guo@guo-ubu:~/eng/ECO-pytorch-master$

hello,I want to set num_segments to 16,I have tried to change
args.num_segments=16,but it's wrong.Can you tell me what should I do?

and the base_model include one segments(input one frame ) or four segment network?
thank you very much