surmenok / gtsrb

Convolutional Neural Network for German Traffic Sign Recognition Benchmark

Overview

A convolutional neural network for German traffic sign image classification.

Dataset

German Traffic Sign Recognition Dataset (GTSRB) is an image classification dataset.
The images are photos of traffic signs. The images are classified into 43 classes. The training set contains 39209 labeled images and the test set contains 12630 images. Labels for the test set are not published.
See more details here.

Model

ResNet-34 pretrained on ImageNet dataset, then finetuned on GTSRB dataset.

Deep Learning Libraries

fastai with PyTorch backend.

Metrics

The model achieved 99.22% accuracy on the validation set (random 20% subset of the training dataset).

gtsrb's Issues

name 'ModelDataLoader' is not defined

env: Ubuntu 16.04 x64, Python 3.6.6, Pytorch 0.2.0, web.py 0.40.dev0

NameError Traceback (most recent call last)
in ()
3 aug_tfms = [RandomRotate(20), RandomLighting(0.8, 0.8)]
4 tfms = tfms_from_model(arch, sz, aug_tfms=aug_tfms, max_zoom=1.2)
----> 5 data = ImageClassifierData.from_paths(path, tfms=tfms, test_name='test', bs=bs)
6
7 ims = np.stack([get_augs() for i in range(6)])

in from_paths(cls, path, bs, tfms, trn_name, val_name, test_name, num_workers)
60 test_fnames = read_dir(path, test_name) if test_name else None
61 datasets = cls.get_ds(FilesIndexArrayDataset, trn, val, tfms, path=path, test=test_fnames)
---> 62 return cls(path, datasets, bs, num_workers, classes=trn[2])
63
64 @classmethod

in init(self, path, datasets, bs, num_workers, classes)
6 self.get_dl(ds,shuf) for ds,shuf in [
7 (trn_ds,True),(val_ds,False),(fix_ds,False),(aug_ds,True),
----> 8 (test_ds,False),(test_aug_ds,False)
9 ]
10 ]

in (.0)
4 self.path,self.bs,self.num_workers,self.classes = path,bs,num_workers,classes
5 self.trn_dl,self.val_dl,self.fix_dl,self.aug_dl,self.test_dl,self.test_aug_dl = [
----> 6 self.get_dl(ds,shuf) for ds,shuf in [
7 (trn_ds,True),(val_ds,False),(fix_ds,False),(aug_ds,True),
8 (test_ds,False),(test_aug_ds,False)

in get_dl(self, ds, shuffle)
13 def get_dl(self, ds, shuffle):
14 if ds is None: return None
---> 15 return ModelDataLoader.create_dl(ds, batch_size=self.bs, shuffle=shuffle,
16 num_workers=self.num_workers, pin_memory=False)
17

NameError: name 'ModelDataLoader' is not defined

What version of fastai and torch should we use?

Thx~

FileNotFoundError: test folder doesn't exist or is empty

Notebook cell 👇🏼

bs = 256

aug_tfms = [RandomRotate(20), RandomLighting(0.8, 0.8)]
tfms = tfms_from_model(arch, sz, aug_tfms=aug_tfms, max_zoom=1.2)
data = ImageClassifierData.from_paths(path, tfms=tfms, test_name='test', bs=bs)

ims = np.stack([get_augs() for i in range(6)])
plots(ims, rows=2)

Too long output of the 3rd cell

It is hard to view notebook in nbviewer as there is too long output of the third cell. Possibly it is better to add -q in unzip command.