search

dougsm / ggcnn

Generative Grasping CNN from "Closing the Loop for Robotic Grasping: A Real-time, Generative Grasp Synthesis Approach" (RSS 2018)
BSD 3-Clause "New" or "Revised" License
499 stars 140 forks source link

evaluation.calculate_iou_match() is something wrong？ #23

Closed Cheniitt closed 4 years ago

Cheniitt commented 4 years ago

when training the ggcnn2, validating is always zero, but the loss is Decreasing.

Cheniitt commented 4 years ago

image

dougsm commented 4 years ago

g'day @Cheniitt

It definitely looks like there's something strange going on. Can you provide more details? What dataset are you using? What was the command you ran? If you run it with the --vis flag does it look like the network is learning properly?

Cheniitt commented 4 years ago

thanks for your reply @dougsm I used the jacquard dataset to train ggcnn2, run the command python train_ggcnn.py --description training_example2 --network ggcnn2 --dataset jacquard --dataset-path <Path To Dataset> --vis, when training with the flag --vis, it looks the network is learning properly.

Cheniitt commented 4 years ago

I found in https://github.com/dougsm/ggcnn/blob/ad48bc5f768fe0a9ba9fd47729638e0aed46e47b/utils/dataset_processing/grasp.py#L269 Whenever it always returns 0.