Thinklab-SJTU / R3Det_Tensorflow

Code for AAAI 2021 paper: R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object
Apache License 2.0
540 stars 122 forks source link

test_dota_r3det_plusplus_ms.py not working #55

Closed s-shanky-k closed 4 years ago

s-shanky-k commented 4 years ago

I have loaded the pre-trained model in the output/trained_weights as shown below

trained weight struct

I have also loaded the pre-trained weights in the data/pretrained_weights as shown below

Pre-weights models tree struct

I have also modified the cfgs.py code with the cfgs_res152_dota_r3det_plusplus_v1.py code.

Also as per the comments in previous similar issues the code is modified as: image

I execute the code using python test_dota_r3det_plusplus_ms.py --test_dir='../PATH/TO/TEST/DIR/' --gpus=0 -s

The model runs, but the output images have no bounding boxes. Am I doing something wrong?

yangxue0827 commented 4 years ago

It may be that a lot of bugs have been fixed in the code recently, causing the previously trained weights to not be used normally, so you may need to retrain by yourself. @rampage-rider