def cornet_s10rep_brainmodel():
activations_model = cornet('CORnet-S')
old_times = activations_model._model.IT.times
new_times = 10
activations_model._model.IT.times = new_times
size_12 = activations_model._model.IT.norm1_0.num_features
size_3 = activations_model._model.IT.norm3_0.num_features
for t in range(old_times, new_times):
setattr(activations_model._model.IT, f'norm1_{t}',
nn.BatchNorm2d(size_12))
setattr(activations_model._model.IT, f'norm2_{t}',
nn.BatchNorm2d(size_12))
setattr(activations_model._model.IT, f'norm3_{t}',
nn.BatchNorm2d(size_3))
identifier = f'CORnet-S{new_times}rep'
activations_model.identifier = identifier
time_step_size = 10
time_mapping = {
timestep:
(70 + timestep * time_step_size, 70 + (timestep + 1) * time_step_size)
for timestep in range(0, new_times)
}
return CORnetCommitment(
identifier=identifier,
activations_model=activations_model,
layers=['V1.output-t0'] + [
f'{area}.output-t{timestep}'
for area, timesteps in [('V2', range(2)), ('V4', range(4)),
('IT', range(2))] for timestep in timesteps
] + ['decoder.avgpool-t0'],
time_mapping={'IT': time_mapping})
def cornet_r2_brainmodel():
return CORnetCommitment(identifier='CORnet-R2',
activations_model=cornet('CORnet-R2'),
layers=['V1.output-t0'] + [
f'{area}.output-t{timestep}'
for area in ['V2', 'V4', 'IT']
for timestep in range(5)
] + ['avgpool-t0'],
time_mapping={
'V1': {
0: (50, 150)
},
'V2': {
0: (50, 80),
1: (80, 110),
2: (110, 140),
3: (140, 170),
4: (170, 200)
},
'V4': {
0: (70, 110),
1: (110, 140),
2: (140, 170),
3: (170, 200),
4: (200, 250)
},
'IT': {
0: (70, 110),
1: (110, 140),
2: (140, 170),
3: (170, 200),
4: (200, 250)
},
})
def cornet_s_brainmodel():
# map region -> (time_start, time_step_size, timesteps)
time_mappings = CORNET_S_TIMEMAPPING
return CORnetCommitment(
identifier='CORnet-S',
activations_model=cornet('CORnet-S'),
layers=['V1.output-t0'] + [
f'{area}.output-t{timestep}'
for area, timesteps in [('V2', range(2)), ('V4', range(4)),
('IT', range(2))] for timestep in timesteps
] + ['decoder.avgpool-t0'],
time_mapping=_build_time_mappings(time_mappings))
def cornet_s222_brainmodel():
time_start, time_step_size = 70, 100
time_mapping = {
timestep: (time_start + timestep * time_step_size,
time_start + (timestep + 1) * time_step_size)
for timestep in range(0, 2)
}
return CORnetCommitment(
identifier='CORnet-S222',
activations_model=cornet('CORnet-S222'),
layers=['V1.output-t0'] + [
f'{area}.output-t{timestep}'
for area, timesteps in [('V2', range(2)), ('V4', range(4)),
('IT', range(2))] for timestep in timesteps
] + ['decoder.avgpool-t0'],
time_mapping={'IT': time_mapping})
def cornet_s_brainmodel():
# map region -> (time_start, time_step_size, timesteps)
time_mappings = {
'V1': (50, 100, 1),
'V2': (70, 100, 2),
# 'V2': (20, 50, 2), # MS: This follows from the movshon anesthesized-monkey recordings, so might not hold up
'V4': (90, 50, 4),
'IT': (100, 100, 2),
}
return CORnetCommitment(
identifier='CORnet-S',
activations_model=cornet('CORnet-S'),
layers=['V1.output-t0'] + [
f'{area}.output-t{timestep}'
for area, timesteps in [('V2', range(2)), ('V4', range(4)),
('IT', range(2))] for timestep in timesteps
] + ['decoder.avgpool-t0'],
time_mapping=_build_time_mappings(time_mappings))
def cornet_r10rep_brainmodel():
activations_model = cornet('CORnet-R')
new_times = 10
activations_model._model.times = new_times
activations_model.identifier = f'CORnet-R{new_times}'
time_step_size = 10
time_mapping = {
timestep:
(70 + timestep * time_step_size, 70 + (timestep + 1) * time_step_size)
for timestep in range(0, new_times)
}
return CORnetCommitment(identifier=f'CORnet-R{new_times}',
activations_model=activations_model,
layers=['maxpool-t0'] + [
f'{area}.relu3-t{timestep}'
for area in ['block2', 'block3', 'block4']
for timestep in range(new_times)
] + ['avgpool-t0'],
time_mapping={'IT': time_mapping})
def cornet_z_brainmodel():
return CORnetCommitment(
identifier='CORnet-Z',
activations_model=cornet('CORnet-Z'),
layers=[f'{region}.output-t0' for region in ['V1', 'V2', 'V4', 'IT']] +
['decoder.avgpool-t0'],
time_mapping={
'V1': {
0: (50, 150)
},
'V2': {
0: (70, 170)
},
'V4': {
0: (90, 190)
},
'IT': {
0: (100, 200)
},
})
