evaluation result

[saraershadi]

Hi ,

I want to re evaluate your method and see the simmilar results as you reported in the original paper entitled with "Human Pose Estimation with Parsing Induced Learner".
I successfully could run your provided code in LIP dataset but when I run the command
sh run_test.sh
or
CUDA_VISIBLE_DEVICES=0 python main.py --evaluate True --calc-pck True --resume exps/snapshots/pil_lip_best.pth.tar

The results is very far from the results that are reported in the original paper.

By running your code I see it writes that
.
.
.
Testing for LIP dataset: [9997/10000], name: 99681_521075.jpg, cur time: 0.2338, avg time: 0.2340
Testing for LIP dataset: [9998/10000], name: 99807_183413.jpg, cur time: 0.2342, avg time: 0.2340
Testing for LIP dataset: [9999/10000], name: 99938_1719104.jpg, cur time: 0.2381, avg time: 0.2340
Testing for LIP dataset: [10000/10000], name: 99938_471406.jpg, cur time: 0.2355, avg time: 0.2340
[email protected] & Head & Sho. & Elb. & Wri. & Hip & Knee & Ank. & U.Body & Avg.
hpe_with_pil & 0.3 & 0.1 & 0.1 & 0.2 & 0.0 & 3.9 & 4.8 & 0.2 & 1.1

At the final line the results indicates that Avg is very low by running your provided code.

Thank you
Sarah