AttributeError: module 'albumentations' has no attribute 'Sequentials' about keypoint_rcnn_training_pytorch

[schleptech]

Good deal! Got everything to work for me -
so to summarize - I followed your step of
pip install -U albumentations --no-binary qudida,albumentations

Which installed the newer version of albumentations (colab has an old version) and the script also brought in the appropriate opencv-python-headless.
Note: in google colab (which I am not familiar with particularly; I primarily use AWS EC2 instances via SSH) you must restart the runtime!

Also remember to change cocoeval.py everytime you bring the runtime back up.

Working just as you specified. Thanks!

[alexppppp]

You need to install a newer version of albumentations library

[schleptech]

I read that Google colab has an outdated version, so I ran the following at the beginning of the notebook:

!pip install -q -U albumentations
!echo "$(pip freeze | grep albumentations) is successfully installed"

[https://colab.research.google.com/github/albumentations-team/albumentations_examples/blob/colab/tensorflow-example.ipynb#scrollTo=26f58a5d]

However, when I get to that step again, I am still getting the error

AttributeError: module 'albumentations' has no attribute 'Sequential'

[schleptech]

I got past this error by following these instructions:

It has something to do with the version of opencv-python-headless. So I updated that in the beginning and we got past the sequentials error with the following.

!pip install --upgrade --force-reinstall --no-deps albumentations
!pip install opencv-python-headless==4.5.5.62

[alexppppp]

So, the error disappeared, correct?

P.S. This way of update is preferrable (from https://albumentations.ai/docs/getting_started/installation/):
If you already have some OpenCV distribution (such as opencv-python-headless, opencv-python, opencv-contrib-python or opencv-contrib-python-headless) installed in your Python environment, you can force Albumentations to use it by providing the --no-binary qudida,albumentations argument to pip, e.g.
pip install -U albumentations --no-binary qudida,albumentations

[schleptech]

Yes - updated my notebook to reflect your above suggestion.

Now when I get down to step 4, "4. Visualizing a random item from dataset", I get the following error:

SystemError: new style getargs format but argument is not a tuple

Seems to be an opencv error

Full output:
SystemError Traceback (most recent call last)
in ()
51 keypoints_original.append([kp[:2] for kp in kps])
52
---> 53 visualize(image, bboxes, keypoints, image_original, bboxes_original, keypoints_original)

in visualize(image, bboxes, keypoints, image_original, bboxes_original, keypoints_original)
27 for idx, kp in enumerate(kps):
28 image_original = cv2.circle(image_original, tuple(kp), 5, (255,0,0), 10)
---> 29 image_original = cv2.putText(image_original, " " + keypoints_classes_ids2names[idx], (kp), cv2.FONT_HERSHEY_SIMPLEX, 2, (255,0,0), 3, cv2.LINE_AA)
30
31 f, ax = plt.subplots(1, 2, figsize=(40, 20))

`

[alexppppp]

Yes, the same for me

It can be fixed with changing this line:
image_original = cv2.putText(image_original, " " + keypoints_classes_ids2names[idx], (kp), cv2.FONT_HERSHEY_SIMPLEX, 2, (255,0,0), 3, cv2.LINE_AA)

to this line:
image_original = cv2.putText(image_original, " " + keypoints_classes_ids2names[idx], tuple(kp), cv2.FONT_HERSHEY_SIMPLEX, 2, (255,0,0), 3, cv2.LINE_AA)

Update: I've made this change in the original notebook