In this folder, you can find the code to generate the most exciting stimuli using ViV1T (Figure 5 and Supplemental Figure 4). Note that we centre the stimuli to the estimated receptive field of the neuron, see tuning_retinotopy/README.md.
Let us first look at some examples from the paper. Here in particular, we will focus on contextual modulation at the neuron level.
Natural and ViV1T-generated surrounds elicit stronger contextual modulation than gratings (Figure 5A)
| Mouse (FOV) |
Neuron | Most exciting centre |
Most exciting surrounds | ||
|---|---|---|---|---|---|
| Grating video |
Natural video |
Generated video |
|||
| L | 050 |  |
 |
 |
 |
| N | 054 |  |
 |
 |
 |
| N | 059 |  |
 |
 |
 |
| L | 071 |  |
 |
 |
 |
Quantifying the responses of many neurons to these stimuli, we found that natural surrounds typically allow for more excitation than grating surrounds, and that ViV1T-generated surrounds allow for even more.
| Mouse (FOV) |
Neuron | Most exciting centre |
Most exciting surrounds | |||
|---|---|---|---|---|---|---|
| Natural image | Natural video | Generated image | Generated video | |||
| L | 003 |  |
 |
 |
 |
 |
| N | 054 |  |
 |
 |
 |
 |
| N | 059 |  |
 |
 |
 |
 |
| L | 071 |  |
 |
 |
 |
 |
All of these were generated using the code in this folder:
- estimate_neuron_reliability.py estimate the neuron reliability before predicting and generating its MEIs and MEVs.
- estimate_natural_stimuli_spectrum.py estimate the spatiotemporal power spectrum of the natural movies in the training set. This is needed to compute the KL divergence between generated stimuli and natural stimuli.
- single neuron most-exciting stimuli single_neuron/
- most-exciting grating stimuli single_neuron/grating_stimulus
- predict_center_surround_gratings.py predict centre-surround gratings to find the combination of centre and surround gratings that elicit the strongest response. predict_natural_surround_with_grating_center.py predict and find the natural video surround with the most exciting grating centre fixed that elicits the strongest response.
- most-exciting natural stimuli single_neuron/natural_stimulus
- predict_natural_center.py predict and find the natural centre that elicits the strongest response. predict_natural_surround.py predict and find the natural video surround with the most exciting natural centre fixed that elicits the strongest response.
- generate_center_surround.py generate the most-exciting image (MEI) and/or video (MEV) surround with the most-exciting grating/natural centre fixed.
- most-exciting grating stimuli single_neuron/grating_stimulus
- population most-exciting stimuli population/ mirrors the single neuron folder structure, but predicts and generates stimuli to the population response.