Dataset format: Zones

[VSainteuf]

@VSainteuf @watch24hrs-iiitd, Hello

it is possible to do a new training with a new dataset. I am studying the implementation and I find that the file train_panoptic.py needs the zone parameters for this. What are zones and what type of data are they.

Than you

if mode != "train": with torch.no_grad(): predictions = model( x, batch_positions=dates, pseudo_nms=compute_metrics, heatmap_only=heatmap_only, ) else: zones = y[:, :, :, 2] if config.supmax else None optimizer.zero_grad() predictions = model( x,

Originally posted by @jhonjam in #3 (comment)

[VSainteuf]

Good question!
the Zones tensor represents a mapping between pixel position and ground truth objects. This mapping is used to supervise only one predicted instance per ground truth object.
In the paper we discuss this mapping in the paragraph that follows Eq. 7.
And practically, zones is a HxW long tensor where the value at pixel (i,j) is the instance id of the ground truth object this position is mapped to.

I realise the documentation does not really detail this point, I'll update it.

[VSainteuf]

I also added the script to compute the centerness heatmap and the Zones tensor in the dataset repository. It will probably make things clearer.