Questions about the code

[shangbuhuan13]

ey :)
I am sorry I have another question

in the recon_loss.py

res_vote = res_vote / 6 / bs

all the devisions by 6 is this because of the 6 distances paramters or is this related to the number of categories

and another question

GPV_Pose/losses/recon_loss.py

Line 234 in 5ac3307

if obj_id != 5:

this line why is this not calculated for latop?

[shangbuhuan13]

and same question here how is this selected:

GPV_Pose/network/GPVPose.py

Line 202 in 5ac3307

if prop_bc < FLAGS.aug_bc_pro and (obj_id == 5 or obj_id == 1):

[lolrudy]

1. 6 is the number of the bounding box faces.
2. The category which obj_id==5 is "mug". Since the x-axis of the mug category is shifted by SPD, the ground truth of x-axis is hard to calculated. So we do not calculate the loss on x-axis for mug.
3. We follow the code of FS-Net and only imply box-cage augmentation on "bowl" and "mug". This augmentation preserves the geometry characteristics of these two categories and achieves best performance.

[HannahHaensen]

@lolrudy one question regarding that, but category 5 is the category laptop

GPV_Pose/datasets/load_data.py

Line 74 in 2b9afaf

self.id2cat_name = {'1': 'bottle', '2': 'bowl', '3': 'camera', '4': 'can', '5': 'laptop', '6': 'mug'}

mug is category 'mug': 6 if I read it correctly

thanks in advance

[lolrudy]

The obj_id starts from 0.

GPV_Pose/datasets/load_data.py

Line 264 in 2b9afaf

cat_id = gts['class_ids'][idx] - 1 # convert to 0-indexed