训练车道线和车辆目标检测，该怎么修改config中的default?

Open zhangshengpeng2021 opened this issue 2 years ago • 6 comments

训练车道线和车辆目标检测，该怎么修改config中的default?

Nov 18 '22 08:11 zhangshengpeng2021

You need to add a new flag, this will train only detection branch and lane lines (encoder and drivable area will be freezed)

Change 1

Alternating optimization

_C.TRAIN.SEG_ONLY = False # Only train two segmentation branchs _C.TRAIN.DET_ONLY = False # Only train detection branch _C.TRAIN.ENC_SEG_ONLY = False # Only train encoder and two segmentation branchs _C.TRAIN.ENC_DET_ONLY = False # Only train encoder and detection branch

Single task

_C.TRAIN.DRIVABLE_ONLY = False # Only train da_segmentation task _C.TRAIN.LANE_ONLY = False # Only train ll_segmentation task _C.TRAIN.DET_ONLY = False # Only train detection task

Add this line in config/default.py _C.TRAIN.LANE_AND_DETECTION_ONLY = True

Change2 Add this part in tools/train.py if cfg.TRAIN.LANE_AND_DETECTION_ONLY: logger.info('freeze encoder and Da_Seg heads...') # print(model.named_parameters) for k, v in model.named_parameters(): v.requires_grad = True # train all layers if k.split(".")[1] in Encoder_para_idx + Da_Seg_Head_para_idx : print('freezing %s' % k) v.requires_grad = False

Change3 Add this part in lib/core/loss.py in _forward_impl function : if cfg.LANE_AND_DETECTION_ONLY: lseg_da = 0 * lseg_da