search

facebookresearch / InterHand2.6M

Official PyTorch implementation of "InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image", ECCV 2020
Other
676 stars 92 forks source link

InterHand2.6M_XXX_MANO.json Discard or not? #52

Closed Gzzgz closed 3 years ago

Gzzgz commented 3 years ago

all of the InterHand2.6M_XXX_MANO.json are replaced by InterHand2.6M_XXX_MANO_NeuralAnnot.json ? THX

mks0601 commented 3 years ago

Hi, yes all json files are updated. The new version is better than the old ones, so you can just discard the old ones.

Gzzgz commented 3 years ago

OK THX

But I find InterHand2.6M_XXX_MANO_NeuralAnnot.json is smaller than InterHand2.6M_XXX_MANO.json very much. Is there a lot of missing data?

mks0601 commented 3 years ago

I don't think so.. when did you download the json files? As written in the README, the json files got smaller in 2020.11.26

Gzzgz commented 3 years ago

OK , I think so. the mano labels count like that: all: train:1340326 test:848297 val:379947

I'm just curious about how it's getting smaller. Good Job!

mks0601 commented 3 years ago

Before the update, there were MANO parameters for each camera view and for each time step, which make the number of MANO parameters extremely large. Instead, the updated version has a single MANO parameter for each time step (note that 'each camera view' is removed), which can be transformed to each camera view using camera extrinsic parameters.

Gzzgz commented 3 years ago

OK THX!