Closed KatsuhisaMorita closed 11 months ago
先行研究の手法も参考になりそうですね。
The numbers of collected silhouette images of the crab, lion and hare were 100, 100 and 105, respectively (I bought silhouette images from Deposit Photos (https://jp.depositphotos.com/), and downloaded free images from Silhouette AC (https://www.silhouette-ac.com/index.html), Illust AC (https://www.ac-illust.com), Clipart Library (http://clipart-library.com), Silhouette Design (http://kage-design.com/)). Example images are shown in Fig. 3. Of these 305 images, 240 images (80 images in each class) were used for training of the CNNs and 65 images were used as the test data set. The size of each image was 50×50 pixels and the intensity of the images was normalized between 0 and 1 by dividing each intensity by 255.
そもそもImageNetのsegmentation用に準備されたデータセットを援用できるかもしれない。 https://github.com/LUSSeg/ImageNet-S
231127 data1/ImageNet下にデータセットをダウンロード
231128 ImageNet-S データセットを展開。
https://github.com/LUSSeg/ImageNet-S Image Numbers The ImageNet-S dataset contains 1183322 training, 12419 validation, and 27423 testing images from 919 categories. We annotate 39842 val/test images and 9190 training images with precise pixel-level masks.
具体的に使えそうか、見る。
学習用データセットの準備
ImageNetより加工する
生成AIの利用
その他