Train "sg_grcnn" with batch_size=1

jwyang / graph-rcnn.pytorch

[ECCV 2018] Official code for "Graph R-CNN for Scene Graph Generation"

732 stars 157 forks source link

Train "sg_grcnn" with batch_size=1 #58

Open jgyy4775 opened 5 years ago

jgyy4775 commented 5 years ago

I trained this model with the command "python main.py --config-file configs/sgg_res101_step.yaml --algorithm sg_grcnn". SOLVER: BASE_LR: 5e-3 MAX_ITER: 8000 STEPS: (8000,12000) batch_size : 1

But the result is very strange... Are the steps and batch too small? I wonder why the performance is low.

If not for batch and step did I make a mistake?

August66 commented 4 years ago

hello,do you know why?,i have the same question.

diaoran commented 2 years ago

I have the same question too.Have you solved the problem yet?