andreasveit / triplet-network-pytorch Goto Github PK

Python 100.00%

triplet-network-pytorch's Introduction

A PyTorch Implementation for Triplet Networks

This repository contains a PyTorch implementation for triplet networks.

The code provides two different ways to load triplets for the network. First, it contain a simple MNIST Loader that generates triplets from the MNIST class labels. Second, this repository provides a Triplet Loader that loads images from folders, provided a list of triplets.

Example usage:

$ python train.py

Tracking experiments with Visdom

This repository allows to track experiments with visdom. You can use the VisdomLinePlotter to plot training progress.

If this implementation is useful to you and your project, please also consider to cite or acknowledge this code repository.

triplet-network-pytorch's People

Contributors

Stargazers

Watchers

Forkers

jizhihang benjamesbabala sjtuzhanglj richardhahahaha hyzcn mercileesb fredhuang16 mjchen611 yaoliuoa stalagmite7 twistedmove amena6490 ganghu1993 frank2017 yifei87 pzz2011 santiorlando ngxbac nesrinegrati shushenglee luffydai ramonyeung mahfujau fzulj styanddty sty-dty helenligit shreelock tinyloop yang-zhang subbareddy248 maxton-paycasso wosxcc yypeng9468 qianli123 ankittaxak5717 huyuhuster raghavab1992 ddeeppnneett lillian1082 hareshkarnan noploop swg209 tcrapse cscn89 scarleaf willforcv anirudh257 cwinjet rafis hsdeveloper147 gp1313 douxiaotian wewan fzcz fagan2888 daisydan ljl02521 gcucurull persistentbuilder whz1861 freegliboracle yjzx121 sghoshcvc hukefei yoooo233 wuzhan11 persuelx jasonhai18 fuqianggu gcv9htd juryun ananddb90 meirtz sxndlh emmanuelowusu bertama andybaimqc zyang19 riverhu233 atesnrllh wikiy223 jawaechan aaronponceuv isabella232 shiliang1105 jwang9410 swarchal xie-huan xfp-discerning ylgao1 pangxuejiao sushovanjena axun2001 hitzht

triplet-network-pytorch's Issues

how to test in the test set

I don't understand how to test in the test set

BrokenPipeError:[Errno 32] Broken pipe.

Hi, I use Pytorch to run a triplet network(GPU), but when I got data , there was always a error: BrokenPipeError:[Errno 32] Broken pipe.

I thought it may be something wrong in the following codes：

for batch_idx, (data1, data2, data3) in enumerate(test_loader):
if args.cuda:
data1, data2, data3 = data1.cuda(), data2.cuda(), data3.cuda()
data1, data2, data3 = Variable(data1), Variable(data2), Variable(data3)

Can you give me some suggestions? Thank you so much.

A bug when try to modify your code.

Since I'm new to PyTorch, when I add transforms.CenterCrop(size) to get my dataloader, this error occur

  File "~/triplet-network-pytorch/triplet_data_loader.py", line 73, in __getitem__
    img1 = self.transform(img1)
  File "build/bdist.linux-x86_64/egg/torchvision/transforms.py", line 29, in __call__
    img = t(img)
  File "build/bdist.linux-x86_64/egg/torchvision/transforms.py", line 156, in __call__
    w, h = img.size
TypeError: 'builtin_function_or_method' object is not iterable

If I add transforms.CenterCrop(28) to MNIST_t(*),

train_loader = torch.utils.data.DataLoader(
        MNIST_t('./data', train=True, download=True,
                       transform=transforms.Compose([
                           transforms.ToTensor(),
                           transforms.Normalize((0.1307,), (0.3081,)),
                           # transforms.CenterCrop(28)
                       ])),
        batch_size=args.batch_size, shuffle=True, **kwargs)

It also occur such error. Could you please help me solve it?

Data Load

Hi!

Thank you so much for the code.

I have a dataset for an example: Datasets A,
DatasetA has 50 classes (50 folders). In each folder contains same class images. How can I load the data?

list index out of range

Hi, there is some new error here
how can I handle this error?
line 47, in init
triplets.append((int(line.split()[0]), int(line.split()[1]), int(line.split()[2]))) # anchor, close, far

IndexError: list index out of range

about why dista should larger than distb.

Excuse me.
I have a question that why dista should be larger than distb in --train.py --train() function
I mean that distance between ref and pos should be smaller than ref and neg.
thanks very much!

accuracy of triplet network

Hello, all
the accuracy of the triplet network in the following link is used here. Is it the only way to calculate the accuracy in the triplet network?

triplet-network-pytorch/train.py

Line 251 in 14e6476

def accuracy(dista, distb):

Invalid Index of 0-dim tensor

losses.update(loss_triplet.data[0], data1.size(0))

IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number

No Online Mining

Hi! Great effort on the implementation.
One short coming for this could be that currently this uses Offline Mining.
We could also make the training process more robust by adding Online Mining for the DataLoader.

I would be happy to help with that.

Cheers! :)

doubt: because they sum the embedded and sum with triplet loss ??

triplet-network-pytorch/train.py

Line 150 in 14e6476

loss_embedd = embedded_x.norm(2) + embedded_y.norm(2) + embedded_z.norm(2)

loss_embedd = embedded_x.norm(2) + embedded_y.norm(2) + embedded_z.norm(2)
loss = loss_triplet + 0.001 * loss_embedd

because they sum the embedded and sum with triplet loss ??

Processing Triplet Generation ...
Traceback (most recent call last):
File "train.py", line 257, in
main()
File "train.py", line 61, in main
transforms.Normalize((0.1307,), (0.3081,))
File "/home/qwe/triplet-network-pytorch/triplet_mnist_loader.py", line 44, in init
self.make_triplet_list(n_train_triplets)
File "/home/qwe/triplet-network-pytorch/triplet_mnist_loader.py", line 164, in make_triplet_list
writer.writerows(triplets)
TypeError: a bytes-like object is required, not 'str'

tripletnet.py

Acc = 0

I Finally got where I am wrong. the definition of accuracy seems not compatible with my python version