詢問routines.sh內，train, val, test_probe, test_gallery意義與資料來源

[tony36486]

最近在學習DGD，從results\individually的輸出去追發現輸出在routines.sh內的extract_features()定義。 其中train, val, test_probe, test_gallery的意義與資料來源，無法理解。 原本猜測是在format_資料庫.py中定義的trainval, test_probe, test_gallery，但數目上對不起來。

[Cysu]

These subsets are defined in format_(dataset name).py as you said. Could you please elaborate on where the number mismatches?

When extracting features, we will compute how many iterations do we need, which equals to ceil(#samples / 100). It could extract slightly more sample features to LMDBs, but we will always truncate to the exact number of samples later when converting to numpy.

[tony36486]

從extract_features()去理解train, val, test_probe, testgallery是四個參數。 但在 format(dataset name).py內僅找到trainval, test_probe, test_gallery三個參數被定義。 這是我指的數目上對不起來的意思，解釋不周不好意思。 如果我的理解沒有錯誤，那我想知道trainval如何轉換成train和val。

[Cysu]

When running scripts/make_dbs.sh, it will call tools/make_lists_id_training.py, where we will split the trainval into train and val. Please check the code here.

[tony36486]

thanks for your kindness help.