Warnings when trying to make an ImageDataBunch

Part 1 (2019)
1

I’m trying to make an ImageDataBunch from the training images collected here:

Which are different sizes and rotations and such. But when I run this:

label_pattern = r'/board\d+_(red|green|purple)-(diamond|oval|squiggle)-(empty|filled|hashed)-(single|double|triple).png$'
bunch = ImageDataBunch.from_name_re('.', imagefiles, label_pattern, ds_tfms=get_transforms(max_zoom=1, max_warp=None), size=224, bs=bs, resize_method=ResizeMethod.SQUISH).normalize(imagenet_stats)

I get a boat ton of warnings like:

/homedir/.local/share/virtualenvs/jupyter-5S1mKFjG/lib/python3.7/site-packages/torch/nn/functional.py:2693: UserWarning: Default grid_sample and affine_grid behavior will be changed to align_corners=False from 1.4.0. See the documentation of grid_sample for details.
  warnings.warn("Default grid_sample and affine_grid behavior will be changed "

What am I doing wrong?

3 Likes
2

It’s not your code, it’s a change in PyTorch 1.3.0 that fastai hasn’t been updated for yet.

3 Likes
3

As noted in the github issue for it you can suppress the message with:

import warnings
warnings.filterwarnings("ignore", category=UserWarning, module="torch.nn.functional")
11 Likes
4

I’m running into the same problem and filtering this warning definitely cleans up the notebook, but I can’t tell if something else is going on too – I’m running into a weird problem with jupyter stalling on the creation of an ImageDataBunch in a notebook that previously worked fine (as of a few days ago).

My code is quite simple:

from fastai.vision import *
from fastai.metrics import error_rate
bs = 64 # batch size at 64
base_dir = "/pedfast-pv/code/experiment-1/annotated/DS1/"
labeled_images = get_image_files(base_dir + "labeled/") # hello helpful library!
labeled_images[:5]
np.random.seed(1010102)
pat = r'/([^/]+)_.*.png$'
data = ImageDataBunch.from_name_re(base_dir+"labeled/", labeled_images, pat, ds_tfms=get_transforms(), size=224, bs=bs
                                  ).normalize(imagenet_stats)

All of that runs pretty smoothly.

data.show_batch(rows=3, figsize=(7,6))

^^^ that line causes the notebook to stall, for a VERY long time and eventually finishes and it results in the symptom below:

The only symptom I can see is that ZMQbg/1 on the server running the notebook is pegged at nearly 100% CPU utilization and has a huge alleged allocation of memory:

Tasks:  24 total,   2 running,  22 sleeping,   0 stopped,   0 zombie
%Cpu(s):  2.4 us,  0.5 sy,  0.0 ni, 97.1 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
KiB Mem : 79123404+total, 70943737+free, 11638572 used, 70158128 buff/cache
KiB Swap:        0 total,        0 free,        0 used. 77615289+avail Mem

   PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
 71070 root      20   0 12.143g 3.261g 427964 R 100.3  0.4   5:45.58 ZMQbg/1
 70999 root      20   0 9540924 1.091g 161880 S   1.7  0.1   0:38.47 ZMQbg/1
 69892 root      20   0  251244  79064  15800 S   1.3  0.0   0:12.80 ZMQbg/1
 15322 root      20   0   24748   4628   3276 S   1.0  0.0   2:35.44 htop```
5

Hey @djmax,
I’m not sure if you got this worked out, but I just got a similar issue and kinda “figured it out”.
I’m totally new in every sense of the word but thought maybe I share this.
While doing the lesson 1 on Colab, I got the error
colab warnings.warn("Default grid_sample and affine_grid behavior will be changed
while making the ImageDataBunch, my browser froze, and everything stopped. As stated by @TomB (Thank you), it seems related to the version of pytorch.
I solve it by !pip install -q torch==1.0.0 torchvision to downgrade the torch version. I don’t know if this is healthy, but everything worked fine afterwards.

6

The import warnings and running warning.filterwarning() worked for me.

7

Thank you so much sir, really appreciate your help. thank you.

8

Thanks!

9

This helped immensely… I was getting crazy over this error :slight_smile:

10

Thanks Tom, brilliant!