Protocol

Allen Visual Behavior change detection

Head-fixed mice view a stream of natural images flashed at a fixed cadence (250 ms image, 500 ms gray) and must lick within a short response window after the image identity changes. A subset of trials are catch trials in which the image does not change, and the mouse must withhold licking. The task is the standardized behavioral protocol used by the Allen Institute Visual Behavior 2P and Visual Behavior Neuropixels datasets.

Task structure

Stimulus, choice, feedback, and training as declared in the protocol record.

Stimulus

evidence_schedule
A repeating sequence of identical natural-image flashes is shown until a pseudo-randomly scheduled change time, at which point one image is replaced by a different image drawn from the same image set. On catch trials the scheduled change does not occur and the mouse must withhold licking through the response window.
evidence_type
change-detection
modalities
visual
notes
The relevant operational evidence is a categorical image transition, not a scalar contrast or coherence value. Image identity, change time, and catch-trial flag are preserved as task variables.
units
flashes, seconds
variables
pre-change image identity, change image identity, flashes since trial start, stimulus change indicator

Choice

action_mapping
Licking the spout within the response window after a stimulus change is a go response (hit). Withholding licking through the response window is a no-go response (miss on change trials, correct rejection on catch trials). Licking before the change window is an aborted trial.
alternatives
go, withhold
choice_type
go-no-go
notes
—
response_modalities
lick

Feedback

feedback_type
reward
notes
—
penalty
No active punishment beyond reward omission and trial abort on early licks.
reward
Sweetened water for hits.

Training

notes
Allen Visual Behavior follows a fixed multi-stage curriculum; only the final stage is treated as the protocol-of-record for harmonized analyses.
stages
habituation, go-only static gratings, go-only natural images, full change detection

Trial timing

  1. image_flash
    250 ms

    One natural image is shown.

  2. gray_interval
    500 ms

    Gray screen between flashes.

  3. change_window
    response-limited

    Response window after a scheduled change time.

  4. feedback
    outcome-dependent

Vertical slices

Expected analyses

  • hit rate, miss rate, false-alarm rate, correct-rejection rate
  • d-prime by image change pair
  • lick-latency distribution after change
  • per-image-pair detectability table

Interpretive claims

  • visual change detection
    Confidence: high · ref.garrett-2023-visual-behavior

    The change-detection paradigm probes detection of a categorical transition. Reading it as a measure of generic perceptual sensitivity conflates change detection with novelty, image-set learning, and task-rule familiarity.

Open questions

  • Capture image-set membership (training vs. novel) in slice-level analyses.
  • Decide whether auto-rewarded trials should be retained or filtered in canonical rows.
  • Add explicit VBN neural/probe/channel/unit joins without mixing them into behavioral trial harmonization.

Papers

Findings

Curves attached to this protocol on shared signed-evidence axes.

See all findings overlaid on shared axes.