mks0601 / 3DMPPE_ROOTNET_RELEASE

Official PyTorch implementation of "Camera Distance-aware Top-down Approach for 3D Multi-person Pose Estimation from a Single RGB Image", ICCV 2019
MIT License
477 stars 65 forks source link

Cannot get reported AP on MuPoTS dataset #22

Closed lhyfst closed 4 years ago

lhyfst commented 4 years ago

I downloaded the bbox from https://drive.google.com/drive/folders/1y0NM6dBbfQgOTRQ6t-V6qdvIt2J6rndS and downloaded the pre-trained model from https://drive.google.com/drive/folders/1nQfOIgc7_AG5xPAO-vtG_L0WxdOelxet. But I could not reproduce the AP score(31.0) reported in README.md

My result is

~/3DMPPE_ROOTNET_RELEASE/main$ python test.py --gpu 1 --test_epoch 18
>>> Using GPU: 1
07-14 17:00:48 Creating dataset...
loading annotations into memory...
Done (t=1.49s)
creating index...
index created!
Get bounding box from ../data/MuPoTS/bbox/bbox_mupots_output.json
07-14 17:00:58 Load checkpoint from ../output/muco_coco/model_dump/snapshot_18.pth.tar
07-14 17:00:58 Creating graph...
100%|████████████████████████████████████| 2163/2163 [16:05<00:00,  2.24it/s]
Evaluation start...
Test result is saved at ../output/muco_coco/result/bbox_root_mupots_output.json
loading annotations into memory...
Done (t=1.74s)
creating index...
index created!
AP_root: 0.1514230417627361
mks0601 commented 4 years ago

image

This is what I've got just before. The bbox link you provided contains root joint coordinates and extended bbox. You should use https://drive.google.com/drive/u/1/folders/1LVpqm6WoiaH0320kKDQDxXrHZs6xHnwt

lhyfst commented 4 years ago

Thanks for reply :)

However, the link you provide needs access

67f1f30d13f37dd869eae20400b56e1

mks0601 commented 4 years ago

Sorry I forgot to share this in public. You can use this link or the link at README.

lhyfst commented 4 years ago

Thanks:)