Bias-Compensated Integral Regression for Human Pose Estimation
In human and hand pose estimation, heatmaps are a crucial intermediate representation for a body or hand keypoint. Two popular methods to decode the heatmap into a final joint coordinate are via an argmax, as done in heatmap detection, or via softmax and expectation, as done in integral regression. Integral regression is learnable end-to-end, but has lower accuracy than detection. This paper uncovers an induced bias from integral regression that results from combining the softmax and the expectation operation. This bias often forces the network to learn degenerately localized heatmaps, obscuring the keypoint's true underlying distribution and leads to lower accuracies. Training-wise, by investigating the gradients of integral regression, we show that the implicit guidance of integral regression to update the heatmap makes it slower to converge than detection. To counter the above two limitations, we propose Bias Compensated Integral Regression (BCIR), an integral regression-based framework that compensates for the bias. BCIR also incorporates a Gaussian prior loss to speed up training and improve prediction accuracy. Experimental results on both the human body and hand benchmarks show that BCIR is faster to train and more accurate than the original integral regression, making it competitive with state-of-the-art detection methods.
Medienart: |
E-Artikel |
---|
Erscheinungsjahr: |
2023 |
---|---|
Erschienen: |
2023 |
Enthalten in: |
Zur Gesamtaufnahme - volume:45 |
---|---|
Enthalten in: |
IEEE transactions on pattern analysis and machine intelligence - 45(2023), 9 vom: 05. Sept., Seite 10687-10702 |
Sprache: |
Englisch |
---|
Beteiligte Personen: |
Gu, Kerui [VerfasserIn] |
---|
Links: |
---|
Themen: |
---|
Anmerkungen: |
Date Completed 08.08.2023 Date Revised 10.08.2023 published: Print-Electronic Citation Status MEDLINE |
---|
doi: |
10.1109/TPAMI.2023.3264742 |
---|
funding: |
|
---|---|
Förderinstitution / Projekttitel: |
|
PPN (Katalog-ID): |
NLM355228211 |
---|
LEADER | 01000naa a22002652 4500 | ||
---|---|---|---|
001 | NLM355228211 | ||
003 | DE-627 | ||
005 | 20231226063915.0 | ||
007 | cr uuu---uuuuu | ||
008 | 231226s2023 xx |||||o 00| ||eng c | ||
024 | 7 | |a 10.1109/TPAMI.2023.3264742 |2 doi | |
028 | 5 | 2 | |a pubmed24n1184.xml |
035 | |a (DE-627)NLM355228211 | ||
035 | |a (NLM)37018104 | ||
040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
041 | |a eng | ||
100 | 1 | |a Gu, Kerui |e verfasserin |4 aut | |
245 | 1 | 0 | |a Bias-Compensated Integral Regression for Human Pose Estimation |
264 | 1 | |c 2023 | |
336 | |a Text |b txt |2 rdacontent | ||
337 | |a ƒaComputermedien |b c |2 rdamedia | ||
338 | |a ƒa Online-Ressource |b cr |2 rdacarrier | ||
500 | |a Date Completed 08.08.2023 | ||
500 | |a Date Revised 10.08.2023 | ||
500 | |a published: Print-Electronic | ||
500 | |a Citation Status MEDLINE | ||
520 | |a In human and hand pose estimation, heatmaps are a crucial intermediate representation for a body or hand keypoint. Two popular methods to decode the heatmap into a final joint coordinate are via an argmax, as done in heatmap detection, or via softmax and expectation, as done in integral regression. Integral regression is learnable end-to-end, but has lower accuracy than detection. This paper uncovers an induced bias from integral regression that results from combining the softmax and the expectation operation. This bias often forces the network to learn degenerately localized heatmaps, obscuring the keypoint's true underlying distribution and leads to lower accuracies. Training-wise, by investigating the gradients of integral regression, we show that the implicit guidance of integral regression to update the heatmap makes it slower to converge than detection. To counter the above two limitations, we propose Bias Compensated Integral Regression (BCIR), an integral regression-based framework that compensates for the bias. BCIR also incorporates a Gaussian prior loss to speed up training and improve prediction accuracy. Experimental results on both the human body and hand benchmarks show that BCIR is faster to train and more accurate than the original integral regression, making it competitive with state-of-the-art detection methods | ||
650 | 4 | |a Journal Article | |
650 | 4 | |a Research Support, Non-U.S. Gov't | |
700 | 1 | |a Yang, Linlin |e verfasserin |4 aut | |
700 | 1 | |a Mi, Michael Bi |e verfasserin |4 aut | |
700 | 1 | |a Yao, Angela |e verfasserin |4 aut | |
773 | 0 | 8 | |i Enthalten in |t IEEE transactions on pattern analysis and machine intelligence |d 1979 |g 45(2023), 9 vom: 05. Sept., Seite 10687-10702 |w (DE-627)NLM098212257 |x 1939-3539 |7 nnns |
773 | 1 | 8 | |g volume:45 |g year:2023 |g number:9 |g day:05 |g month:09 |g pages:10687-10702 |
856 | 4 | 0 | |u http://dx.doi.org/10.1109/TPAMI.2023.3264742 |3 Volltext |
912 | |a GBV_USEFLAG_A | ||
912 | |a GBV_NLM | ||
951 | |a AR | ||
952 | |d 45 |j 2023 |e 9 |b 05 |c 09 |h 10687-10702 |