search

geopavlakos / object3d

Code for "6-DoF Object Pose from Semantic Keypoints", ICRA 2017
Other
98 stars 31 forks source link

Training on custom dataset - specific object instance #2

Closed rohanpsingh closed 5 years ago

rohanpsingh commented 6 years ago

Hi @geopavlakos

Thanks for sharing this code with us. But can you tell me approximately how many images were used for training the network in the case of the gas canister? if the number of train images was 150 (as mentioned in the paper), how many epochs were needed to achieve sufficient accuracy?

I am using about 150 images to train it for a single object instance, but the accuracy remains low (less than 0.10) even after running for 100 epochs!

Thanks!