We use profile to customize 4KAgent in different restoration tasks. Here we preset the set up of profile in 4KAgent:
Name: 4KAgent_Profile_Example
PerceptionAgent: llama_vision # [llama_vision, depictqa]
PerceptionAgent_Seed: 1994 # used when PerceptionAgent is llama_vision
Reflection: hpsv2+metric # [hpsv2, hpsv2+metric]
Upscale4K: True # 'Upscale to 4K'
require_sr_size: 300 # if max(H, W) < require_sr_size, append 'super-resolution' for planning, enable when ScaleFactor is null and Upscale4K is false
ScaleFactor: null # [2, 4, 8, 16], higher priority than Upscale4K
RestoreOption: null # Explicity sets the restoration task(s) to be applied. (e.g., 'super-resolution', 'denoise+dehaze')
FaceRestore: True # enable / disable face restoration module in 4KAgent
Brightening: False # enable / disable brightening task in 4KAgent
OldPhotoRestoration: False # enable / disable old photo restoration module in 4KAgent
RestorePerference: Perception # [Fidelity, Perception]
Fast4K: True # enable / disable Fast4K mode in 4KAgent
Fast4kSideThres: 1024 # size threshold when enable Fast4K mode
User_Define: False # Indicates whether to enable a user-specified plan instead of relying on VLM perception
User_Define_Plan: None # Specifies the designated plan, if explicitly provided. (e.g., ["denoising", "super-resolution", "super-resolution"])
with_rollback: True # Whether to trigger rollback in the systemAll these settings have default value in 4KAgent (L128 ~ L171 in the4kagent_pipeline.py).
We evaluate 4KAgent in different tasks under different profiles, we conclude their details and naming convention here:
| Profile Nickname | Perception Agent | Upscale to 4K | Scale Factor | Restore Option | Face Restore | Brightening | Restore Preference |
|---|---|---|---|---|---|---|---|
| Gen4K-P | DepictQA | True | None | None | True | False | Perception |
| Gen4K-F | DepictQA | True | None | None | True | False | Fidelity |
| Aer4K-P | Llama-3.2-Vision | True | None | None | False | False | Perception |
| Aer4K-F | Llama-3.2-Vision | True | None | None | False | False | Fidelity |
| ExpSR-s4-P | Llama-3.2-Vision | False | 4 | super-resolution | False | False | Perception |
| ExpSR-s4-F | Llama-3.2-Vision | False | 4 | super-resolution | False | False | Fidelity |
| ExpSR-s2-F | Llama-3.2-Vision | False | 2 | super-resolution | False | False | Fidelity |
| ExpSR-s8-F | Llama-3.2-Vision | False | 8 | super-resolution | False | False | Fidelity |
| GenSR-s4-P | DepictQA | False | 4 | None | False | False | Perception |
| AerSR-s4-P | Llama-3.2-Vision | False | 4 | None | False | False | Perception |
| AerSR-s4-F | Llama-3.2-Vision | False | 4 | None | False | False | Fidelity |
| GenMIR-P | DepictQA | False | 4 | None | False | True | Perception |
| ExpSRFR-s4-P | Llama-3.2-Vision | False | 4 | super-resolution | True | False | Perception |
| GenSRFR-s4-P | DepictQA | False | 4 | None | True | False | Perception |
Profile naming convention: We combine restoration type, restoration task, and restoration preference to construct the profile name. For example, Gen indicates a General image, 4K indicates "Upscale to 4K" on, and P indicates to restore the image with high Perceptual quality. Aer indicates Aerial image, Exp corresponds to Explicit setting, indicating that the profile has explicitly set the restoration task (e.g., SR, which indicates Super-Resolution). MIR indicates Multiple-Degradation Image Restoration. FR indicates Face Restoration. s4 indicates to upscale the image by a scale factor of 4.