Visualising the evaluation results about learnable-triangulation-pytorch HOT 10 CLOSED

Hi, @Samleo8!
All the visualization tools are collected here: https://github.com/karfly/learnable-triangulation-pytorch/blob/master/mvn/utils/vis.py.

from learnable-triangulation-pytorch.

Also, maybe you will find the ground truth annotations viewer useful.

from learnable-triangulation-pytorch.

Is there documentation on how data in the outputted pickle file looks/is organized?
It seems that in order to use the vis.py tool I need to know how to unpack the data ?

For now, I can see that the dictionary contains something in the form of

"keypoints_3d": (some 3d numpy array?)
"indexes": (??)

May I know what these represent in more detail? Thank you!

from learnable-triangulation-pytorch.

Several other questions as I look through the code of vis.py:

def visualize_batch(images_batch, heatmaps_batch, keypoints_2d_batch, proj_matricies_batch,
                    keypoints_3d_batch_gt, keypoints_3d_batch_pred,

It seems visualize_batch is the most likely function for use for my purpose?

However, I do not know what to pass in for images_batch and keypoints_3d_batch_gt since the pickle file does not seem to contain any information.

Moreover, I am unsure if directly passing in the unpickled data from keypoints_3d_batch_pred is correct, as it might not be in the correct format required by the function?

Thank you and greatly appreciate your help!

from learnable-triangulation-pytorch.

@Samleo8, do you mean pickle file from this issue?

from learnable-triangulation-pytorch.

@karfly No I meant the results.pkl file that resulted from running train.py --eval and is in

logs/eval_human3.6m.../checkpoints/...

from learnable-triangulation-pytorch.

@karfly @shrubb @Danivilanova

In a sense, the issue is "resolved" because I can visualise the results through tensorboard.

However, because I have no idea how the results are being formatted/the way to parse them, I don't have a way of using the evaluation results for anything meaningful, nor applying it directly to another dataset of my own? Is there a way to resolve this?

Thank you!

from learnable-triangulation-pytorch.

The structure of the results.pkl file is fully determined by the code. Here (https://github.com/karfly/learnable-triangulation-pytorch/blob/master/train.py#L353) you can find, how we dump the results-dict. The results-dict is filled with data some lines of code above.

from learnable-triangulation-pytorch.

I'm sorry, but I still don't quite understand how to use the visualization tool. Do I have to write a separate program that calls these functions? And I have the same question as @Samleo8 regarding the parameters of the visualization function.

from learnable-triangulation-pytorch.

Hi, @davidmaeurer!
We don't have a standalone visualization tool, so you should write a separate program to visualize your data. In train.py you can find example on how to visualize a batch.

from learnable-triangulation-pytorch.