Skip to content

推理耗时与论文给出的数据存在较大出入 #13

@zisang0210

Description

@zisang0210

在mmdet/models/detectors/ae_textspotter.py第465行插入代码

with mmcv.Timer():
    text_det_bboxes, text_det_labels = self.simple_test_text_bboxes(
        x, img_meta, text_proposal_list, self.test_cfg.text_rcnn, rescale=rescale)
    rects = self.simple_test_text_mask(x, img_meta, text_det_bboxes, text_det_labels, rescale=rescale)

执行命令
python tools/rects_test.py --local_rank=0 local_configs/rects_ae_textspotter_lm_r50_1x.py work_dirs/rects_ae_textspotter_lm_r50_1x/rects_ae_textspotter_lm_r50_1x.pth --launcher none --json_out results.json
得到TDM(text detection module)在单块Tesla V100 GPU上的平均耗时是1306ms,与论文中所说在1080Ti GPU上的平均耗时是440ms有较大出入,且整体推理速度是0.66fps,与论文中1.3fps存在较大差距。除此之外,根据results.json计算评价指标时可以复现README.md中的结果。请问这是什么原因?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions