Created by: XiminLin

我的模型参数:

YOLOv3: backbone: ResNet yolo_head: YOLOv3Head ResNet: norm_type: bn freeze_norm: false norm_decay: 0. variant: d depth: 50 dcn_v2_stages: [5] feature_maps: [3, 4, 5] YOLOv3Head: anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]] anchors: [[19, 29], [28, 20], [25, 40], [31, 47], [36, 37], [41, 26], [47, 66], [48, 33], [67, 53]] norm_decay: 0. yolo_loss: YOLOv3Loss nms: background_label: -1 keep_top_k: 100 nms_threshold: 0.45 nms_top_k: 1000 normalized: false score_threshold: 0.01 YOLOv3Loss: # batch_size here is only used for fine grained loss, not used # for training batch_size setting, training batch_size setting # is in configs/yolov3_reader.yml TrainReader.batch_size, batch # size here should be set as same value as TrainReader.batch_size batch_size: 8 ignore_thresh: 0.7 # default is 0.7 label_smooth: true

我在 use_fine_grained_loss 的设置如下:

- !Gt2YoloTarget anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]] anchors: [[10, 13], [16, 30], [33, 23], [30, 61], [62, 45], [59, 119], [116, 90], [156, 198], [373, 326]] downsample_ratios: [32, 16, 8]

我用 pretrained weights 得到的 output (use_fine_grained_loss=true):

2020-08-20 11:21:00,054-INFO: iter: 0, lr: 0.000100, 'loss_xy': '11.305380', 'loss_wh': '5.507521', 'loss_obj': '66.596405', 'loss_cls': '0.698010', 'loss': '84.107315', time: 0.000, eta: 0:00:00 2020-08-20 11:21:24,467-INFO: iter: 1, lr: 0.000100, 'loss_xy': '11.817931', 'loss_wh': '5.780123', 'loss_obj': '63.548248', 'loss_cls': '0.411505', 'loss': '81.557808', time: 12.749, eta: 0:00:50

我用 pretrained weights 得到的 output (use_fine_grained_loss=false):

2020-08-20 11:42:41,885-INFO: iter: 0, lr: 0.000100, 'loss': '13.517960', time: 0.000, eta: 0:00:00 2020-08-20 11:43:07,319-INFO: iter: 1, lr: 0.000100, 'loss': '14.360963', time: 12.793, eta: 0:00:51

谢谢