Keumgang Cha

> Could you provide any config file which use this neck? ```python model = dict( type='ImageClassifier', backbone=dict( type='ResNet', depth=50, num_stages=4, out_indices=(0,1,2,3,), style='pytorch', ), neck=dict( type='MultiheadAttentionPooling', in_channels=(256, 512, 1024, 2048), num_heads=(8,...

However, mmdet only provides swin transformer not vision transformer. https://github.com/open-mmlab/mmdetection/tree/master/mmdet/models/backbones

I think you may put command like ``` sh start.sh ``` But I consider that you have to command like ``` nohup sh start.sh ``` Thank you.

I think you maybe not running 17th actor task. Rerun the 17th task by command ``` python trainer_invader.py --num_actors=32 --task=17 --batch_size=32 --queue_size=128 --trajectory=20 --learning_frame=1000000000 --start_learning=0.0006 --end_learning=0.0 --discount_factor=0.99 --entropy_coef=0.05 --baseline_loss_coef=1.0 --gradient_clip_norm=40.0...

I'm sorry to late for answering this question. I think that the first line of start.sh from ``` python trainer_breakout.py --num_actors=32 --task=0 --batch_size=32 --queue_size=128 --trajectory=20 --learning_frame=1000000000 --start_learning=0.0006 --end_learning=0.0 --discount_factor=0.99 --entropy_coef=0.05...

Sorry to late for answering this question. I have this experiment on this computer settings. ``` DRAM 16G CPU 8 core 1080Ti 16G 16 environment parallel ``` It might spend...

> > Basically it's a uniform grid across image coordinates, and each grid centre is used as an anchor, over which the model regresses the deviation of correct bbox. >...