文件數(shù)量（flods）不是原始數(shù)據(jù)集數(shù)量。原始數(shù)據(jù)集84G，解壓后92G。

從中選取少量數(shù)據(jù)方便跑代碼。

原始數(shù)據(jù)集下載連接：

https://bvisionweb1.cs.unc.edu/ILSVRC2017/download-videos-1p39.php

?

This is a mini version of ILSVRC2015 dataset Detailed informations are as follows:Annotations:VID:train:ILSVRC2015_VID_train_0000:1000 folders:many xmlsval:90 folders:many xmlsData:VID:train:ILSVRC2015_VID_train_0000:1000 folders:many framesval:90 folders:many framestest:100 folders:many framessnippets:train:ILSVRC2015_VID_train_0000:1000 folders:many videosval:90 folders:many videostest:100 folders:many videosImageSets:DET_train.txt:pattern: "path + space + 1"e.g. : "train/ILSVRC2015_train_0000/ILSVRC2015_train_00005005 1"VID_train_15frames.txt:pattern: "path + space + 1 + space + frameID + space + total_frames"e.g. : "train/ILSVRC2015_VID_train_0000/ILSVRC2015_train_00000000 1 10 300"note: select 15 frames from total frames of the video, frameID is the index of those frames(15).VID_train_every10frames.txt:pattern: "path + space + 1 + space + frameID + space + total_frames"e.g. : "train/ILSVRC2015_VID_train_0000/ILSVRC2015_train_00000000 1 0 300"e.g. : "train/ILSVRC2015_VID_train_0000/ILSVRC2015_train_00001004 1 30 41"note: select one frame every 15 frames of the video, frameID is the index of those frames.VID_val_videos.txt:pattern: "path + space + imageID + space + frameID + space + total_frames"e.g. : "val/ILSVRC2015_val_00000000 1 0 464"e.g. : "val/ILSVRC2015_val_00000004 2092 430 804"e.g. : "val/ILSVRC2015_val_00000004 2465 803 804"e.g. : "val/ILSVRC2015_val_00000005 2466 0 464"note: imageID start from 1; frameID start from 0；ImageID is relative to all images，frameID is not.VID_val_frames.txt:pattern: "path + space + imageID"e.g. : "val/ILSVRC2015_val_00000000/000000 1"e.g. : "val/ILSVRC2015_val_00177001/000144 176126"note: imageID start from 1, this path is accurate to the image.

?

總結(jié)

以上是生活随笔為你收集整理的MEGA 视频目标检测 数据集 ： ILSVRC2015 VID 说明的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。