search

ZhangAoCanada / RADDet

Range-Azimuth-Doppler Based Radar Object Detection
MIT License
168 stars 39 forks source link

Should train.py and train_cart.py share the totally same parameters? #52

Open huiwenXie opened 5 months ago

huiwenXie commented 5 months ago

Should train.py and train_cart.py share the same "number of steps" and the totally same parameters in "TRAIN" : { "if_validate": true, "epochs" : 1000, "batch_size" : 3, "focal_loss_iou_threshold" : 0.3, "log_dir": "./logs/RadarResNet", "learningrate_init": 0.0001, "learningrate_end": 0.00001, "learningrate_decay": 0.96, "learningrate_decay_gap": 3000, "startup_steps": 1000, "learningrate_startup": 0.0001, "warmup_steps": 20000, "validate_start_steps": 2000, "validate_gap": 2000 }, ? Thanks a lot!

taj1290 commented 3 months ago

yes, as i tried the same so the results are ok.

huiwenXie commented 3 months ago

yes, as i tried the same so the results are ok.

Thanks a lot. Here

yes, as i tried the same so the results are ok.

Thank a lot. I want to ask you another question. Can you tell me if the CONFIDENCE LOSS goes down for a while and then goes up in your results? 捕获 Here is the picture of my result.