Fastai v2 chat

fastai fastai dev
1019

I made some progress with show_batch method to work with labels as well. Earlier, I was passing it as title to b.show() but now labels can have their own show definition as well

@typedispatch
def show_batch(x:SiameseImage, y, samples, ctxs=None, max_n=6, nrows=None, ncols=4, figsize=None, **kwargs):
    n = min(x[0].shape[0], max_n)
    if ctxs is None: ctxs = get_grid(n*2, nrows=None, ncols=ncols, figsize=figsize)
    xiter = samples.itemgot(0)
    def _show(b,l,c,show_label=False, **kwargs):
      c = b.show(ctx=c, **kwargs)
      if show_label:  c = l.show(ctx=c, **kwargs)
      return c
    for i in range(2):
      ctxs[i::2] = [_show(b,l,c,i%2==0) for b,l,c,_ in zip(xiter.itemgot(i),samples.itemgot(1),ctxs[i::2], range(n))]
    return ctxs

Now this works with MultiCategoryBlock or any similar custom label definition. The show_label condition (i%2==0) could be changed to m with m images in a tuple and label will be shown on every m^{th} axis

1 Like
1020

It won’t, the transforms are still applied to all subclasses of TensorMask thanks to typedispatch.

I’ve written a very quick introduction to typedispatch here

1 Like
1021

Following a little bit the issue. I am trying to add histogram equialization for the images I am working in adding a line in load_image like;

def load_image_eq(fn, mode=None, **kwargs):
    "Open and load a `PIL.Image` and convert to `mode`"
    im = Image.open(fn, **kwargs)
    im.load()
    im = im._new(im.im)
    im = ImageOps.equalize(im, mask = None)  ###Added to perform histogram equalization
    return im.convert(mode) if mode else im

I am creating a new class for it:

class PILBase_eq(Image.Image, metaclass=BypassNewMeta):
    _bypass_type=Image.Image
    _show_args = {'cmap':'gray'}
    _open_args = {'mode': 'L'}
    @classmethod
    def create(cls, fn:(Path,str,Tensor,ndarray,bytes), **kwargs)->None:
        "Open an `Image` from path `fn`"
        if isinstance(fn,TensorImage): fn = fn.permute(1,2,0).type(torch.uint8)
        if isinstance(fn,Tensor): fn = fn.numpy()
        if isinstance(fn,ndarray): return cls(Image.fromarray(fn))
        if isinstance(fn,bytes): fn = io.BytesIO(fn)
        return cls(load_image_eq(fn, **merge(cls._open_args, kwargs)))

    def show(self, ctx=None, **kwargs):
        "Show image using `merge(self._show_args, kwargs)`"
        return show_image(self, ctx=ctx, **merge(self._show_args, kwargs))

    def __repr__(self): return f'{self.__class__.__name__} mode= {self.mode} size={"x".join([str(d) for d in self.size])}'

class PILImage_eq(PILBase_eq): pass

class TensorImage_eq(TensorImage): _show_args = PILImage_eq._show_args
PILImage_eq._tensor_cls = TensorImage_eq

However, when trying to generate the dataloader I got the following error:

dls = dblock.dataloaders(path, bs=8)
dls.show_batch()

Could not do one pass in your dataloader, there is something wrong in it
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-29-f74dbf2b72eb> in <module>
      1 dls = dblock.dataloaders(path, bs=8)
----> 2 dls.show_batch()

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/core.py in show_batch(self, b, max_n, ctxs, show, **kwargs)
     88 
     89     def show_batch(self, b=None, max_n=9, ctxs=None, show=True, **kwargs):
---> 90         if b is None: b = self.one_batch()
     91         if not show: return self._pre_show_batch(b, max_n=max_n)
     92         show_batch(*self._pre_show_batch(b, max_n=max_n), ctxs=ctxs, max_n=max_n, **kwargs)

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in one_batch(self)
    129     def one_batch(self):
    130         if self.n is not None and len(self)==0: raise ValueError(f'This DataLoader does not contain any batches')
--> 131         with self.fake_l.no_multiproc(): res = first(self)
    132         if hasattr(self, 'it'): delattr(self, 'it')
    133         return res

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastcore/utils.py in first(x)
    174 def first(x):
    175     "First element of `x`, or None if missing"
--> 176     try: return next(iter(x))
    177     except StopIteration: return None
    178 

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in __iter__(self)
     95         self.randomize()
     96         self.before_iter()
---> 97         for b in _loaders[self.fake_l.num_workers==0](self.fake_l):
     98             if self.device is not None: b = to_device(b, self.device)
     99             yield self.after_batch(b)

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/torch/utils/data/dataloader.py in __next__(self)
    343 
    344     def __next__(self):
--> 345         data = self._next_data()
    346         self._num_yielded += 1
    347         if self._dataset_kind == _DatasetKind.Iterable and \

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/torch/utils/data/dataloader.py in _next_data(self)
    383     def _next_data(self):
    384         index = self._next_index()  # may raise StopIteration
--> 385         data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
    386         if self._pin_memory:
    387             data = _utils.pin_memory.pin_memory(data)

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py in fetch(self, possibly_batched_index)
     32                 raise StopIteration
     33         else:
---> 34             data = next(self.dataset_iter)
     35         return self.collate_fn(data)
     36 

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in create_batches(self, samps)
    104         self.it = iter(self.dataset) if self.dataset is not None else None
    105         res = filter(lambda o:o is not None, map(self.do_item, samps))
--> 106         yield from map(self.do_batch, self.chunkify(res))
    107 
    108     def new(self, dataset=None, cls=None, **kwargs):

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in do_batch(self, b)
    125     def create_item(self, s):  return next(self.it) if s is None else self.dataset[s]
    126     def create_batch(self, b): return (fa_collate,fa_convert)[self.prebatched](b)
--> 127     def do_batch(self, b): return self.retain(self.create_batch(self.before_batch(b)), b)
    128     def to(self, device): self.device = device
    129     def one_batch(self):

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in create_batch(self, b)
    124     def retain(self, res, b):  return retain_types(res, b[0] if is_listy(b) else b)
    125     def create_item(self, s):  return next(self.it) if s is None else self.dataset[s]
--> 126     def create_batch(self, b): return (fa_collate,fa_convert)[self.prebatched](b)
    127     def do_batch(self, b): return self.retain(self.create_batch(self.before_batch(b)), b)
    128     def to(self, device): self.device = device

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in fa_collate(t)
     44     b = t[0]
     45     return (default_collate(t) if isinstance(b, _collate_types)
---> 46             else type(t[0])([fa_collate(s) for s in zip(*t)]) if isinstance(b, Sequence)
     47             else default_collate(t))
     48 

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in <listcomp>(.0)
     44     b = t[0]
     45     return (default_collate(t) if isinstance(b, _collate_types)
---> 46             else type(t[0])([fa_collate(s) for s in zip(*t)]) if isinstance(b, Sequence)
     47             else default_collate(t))
     48 

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/fastai2/data/load.py in fa_collate(t)
     45     return (default_collate(t) if isinstance(b, _collate_types)
     46             else type(t[0])([fa_collate(s) for s in zip(*t)]) if isinstance(b, Sequence)
---> 47             else default_collate(t))
     48 
     49 # Cell

~/anaconda3/envs/fastai2/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py in default_collate(batch)
     79         return [default_collate(samples) for samples in transposed]
     80 
---> 81     raise TypeError(default_collate_err_msg_format.format(elem_type))

TypeError: default_collate: batch must contain tensors, numpy arrays, numbers, dicts or lists; found <class '__main__.PILImage_eq'>
1022

I see there’s a # TODO: docs in nbs/07_vision.core.ipynb
I’m a bit unclear on how exactly the documentation should look like. Does something like this work?:

Screenshot 2020-04-05 at 10.22.32 AM
Screenshot 2020-04-05 at 10.22.32 AM1678×1386 262 KB

1023

There’s no need to modify base class for this. Your operation has to be performed on PIL.Image and item level representation of images is PIL.Image only. So you can simply put this in item_tfms like so:

pipe = Pipeline([PILImage.create,  ImageOps.equalize])  #pipeline
ds = Datasets(items, tfms=pipe) #datasets

def HImageBlock():  # For DataBlock
  return TransformBlock(type_tfms=[PILImage.create,ImageOps.equalize], batch_tfms=IntToFloatTensor)
2 Likes
1024

For the documentation, please refer to this topic. We don’t need help for notebooks where we haven’t done the documentation yet.

1 Like
1025

The result of amazing work @boris has been doing for a while means you need to make sure you have the latest fastcore to run fastai2 (as is always good practice). So if you see an error log_args is not defined while trying to use fastai2, make sure you have your editable install of fastcore and are up to date with master.

1 Like
1027

@log_args(but=‘dls,model’)

what is the functionality of log_args ?

thanks

1028

It adds the args you used into self.init_args.
The idea is to easily debug when you have a problem by seeing all the args that have been used.
Also it can be used by logging callbacks (eg WandbCallback) to be automatically logged as config parameters.

4 Likes
1029

It is going to be really sweet once finished, I’m pretty excited about it :slight_smile:

3 Likes
1030

I see this error too and I am installing like this:

pip install git+https://github.com/fastai/fastai2.git
pip install git+https://github.com/fastai/fastcore.git
1031

Solved with next order:

pip install git+https://github.com/fastai/fastcore.git@master
pip install git+https://github.com/fastai/fastai2.git@master

@sgugger maybe it is a good idea to update the FAQ because it has the order that caused me the error. Fastai-v2 FAQ and links (read this before posting please!)

1032

You might be interested in other approach for Siamese. One samples pairs of the same class in the dataset, pack in batches and negatives are mined in the batch. No specific class needed.

I have implemented it for patch descriptor learning:

3 Likes
1033

Btw, is there a way to skip computing validation loss?

1034

You can raiske a CancelValidation error (can’t remember the exact name but you should find it easily) that will skip the validation phase.
If the questions is skipping loss but still computing other metrics, the answer is no however.

1035

I always use

pip install git+https://github.com/fastai/fastcore.git --upgrade
pip install git+https://github.com/fastai/fastai2.git --upgrade

which is the better approach?

1 Like
1036

I don’t know, I just added the branch at the end.

1 Like
1037

Skipping loss, but do val. Ok, I’ll do my callback then

1038

So you’ve written imshow_torch to work with that, that’s pretty cool !

I wonder what would happen if you call _pre_show_batch on your dataloaders :thinking:

1039

I was running lr_find and I see this printed

@log_args had an issue on LabelSmoothingCrossEntropy.__init__ -> missing a required argument: 'self'

It’s not throwing any error though, anything to be concerned about?

1 Like