Deformable-Sketch(Edge)-Detection-Network

Sketch detection network with deformable convolution.
When the input image is given, then model outs it's sketch(edge) version.
I use my custom sketch(edge) dataset instead general benchmark datasets.

Test Envrionment

• NVIDIA A5000 24G
• Train 512 x 512 image
• Modeol params: 6.2M

Results

Aru(BlueArchive)

Shiroko(BlueArchive)

Wakamo(BlueArchive)

Surtr(Arknights)

Amiya(Arknights)

Theresa(Arknights)

Benchmark(BSDS500)

BSDS500

BIPEDv2

Multicue