ILSVRC2015 DET

训练和验证数据集与 ILSVRC2014 相同。共有 456567 张图像用于训练。每个同义词集（类别）的正图像数量范围从 461 到 67513。负图像数量范围从每个同义词集 42945 到 70626。有 20121 个验证图像。所有图像均为 JPEG 格式。通过添加 11142 个新图像来刷新测试数据集。该文件包含 51294 (40152 + 11142) 张图像和 test.txt 文件。测试新文件仅包含 11142 个新图像和 test.txt 文件。

The training and validation datasets are identical to those of ILSVRC2014. A total of 456,567 images are used for training. For each synset (category), the number of positive images ranges from 461 to 67,513, while the number of negative images per synset ranges from 42,945 to 70,626. There are 20,121 validation images, and all images are in JPEG format. The test dataset is refreshed by adding 11,142 new images. This file contains 51,294 (40,152 + 11,142) images along with the test.txt file. The new test file only includes 11,142 new images and the test.txt file.

搜集汇总

背景与挑战

背景概述

ILSVRC2015 DET是一个大规模目标检测数据集，包含超过45万张训练图像和2万张验证图像，覆盖广泛的物体类别如乐器、动物和日常用品，适用于多类别目标检测任务。该数据集延续了ILSVRC2014的训练和验证集，并刷新了测试集以增强评估效果。