loveSnowBest
loveSnowBest
You may have a try. The code should not have this problem since I have used it to generate the database before... it's quite strange... Perhaps I changed it for...
Same question. And the initial learning rate in the paper is 1e-2, while 1e-3 in this repo here. Have you checked if we can enlarge the lr while reducing the...
@dianxin556 have you ever tried one 640x640 trained model? My device is not allowed to train 24 images for 840x840 :(
Hi, I've trained one 640x640 baseline, the result is ==================== Results ==================== Easy Val AP: 0.9270121584603086 Medium Val AP: 0.9234432537094537 Hard Val AP: 0.884572269484029
Any plans about knowledge distillation for object detection, for example, fitnet (https://github.com/TuSimple/simpledet/blob/master/models/KD/README.md), so that users can customize their distillation loss functions based on their requirements.
mmdet3d version is 0.16.0, and I think pytorch version is not a problem, maybe 1.5 ? mmcv-full could be an early version, perhaps 1.4, if it does not work, try...
Hi, currently we do not have a plan to release AutoAlign, since AutoAlignV2 performs much better than it.
You may put a pdb and print the dict name in the example['img_info'], and check the dict value is `filename` or `file_name` or other values. Simply changing this could be...
Thanks for your attention to our work. Currently, we have no plan to release the checkpoints. Perhaps you can train the code directly and open an issue if you meet...