Super-resolution

Task Type

Image Restoration

Description

The Super-resolution model enhances spatial resolution of EM images.
It is trained for low-level structural restoration (from EMDiffuse Dataset) without post-processing.

---

Configuration

name: "Super-resolution"
tasktype: 1
solution:
  - backbone_config: vitl.yaml
    datatypes:
      - 2
      - 3
    img_z: 1
    name: emdiffuse-l
    note: trained on EMDiffuse dataset, Low-level, no post-processing, alpha=0.25, beta=0.25
    weights: weights/superreso/981474_model_199.pt

Model Characteristics

• 2D model (img_z = 1)

• Accepts both 2D and 3D input

• 3D input processed slice-wise

• No post-processing applied