search

SpursLipu / YOLOv3v4-ModelCompression-MultidatasetTraining-Multibackbone

YOLO ModelCompression MultidatasetTraining
GNU General Public License v3.0
445 stars 136 forks source link

kmeans anchors #20

Closed SISTMrL closed 4 years ago

SISTMrL commented 4 years ago

when I use the anchor box clustered from my own data, why map is decreased? use the source code's anchors size, my best map is 0.95, but when I use the anchor size for my own data, the best map just is nearly 0.7

SpursLipu commented 4 years ago

I think it depend on distribution of your dataset. If it is concentrated in a small range, clustering will be necessary. But it's just my empirical thought, you can select best way by trying.

SISTMrL commented 4 years ago

Yes，the region I need to detect is small relative to the whole and adjacent to each other, so i do the kmeans anchors

刘志超

邮箱：15256956711@163.com |

签名由 网易邮箱大师 定制

On 05/29/2020 20:02, SpursLipu wrote:

I think it depend on distribution of your dataset. If it is concentrated in a small range, clustering will be necessary. But it's just my empirical thought, you can select best way by trying.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.