ResNet-269 feature-extracting backbone (no classification head).
Builds a ResNet with the very deep bottleneck topology
[3, 30, 48, 8], an extension of the canonical ResNet design
studied in subsequent deep-residual literature (notably as the
teacher network in MXNet-style training pipelines and as an
inception-resnet baseline). Approximately 102.1M parameters.
Model Size
Parameters
pretrainedbool= FalseReserved for future pretrained-weight loading. Currently
ignored.
**overridesobject= {}Keyword overrides forwarded into
ResNetConfig.Returns
ResNetBackbone with the ResNet-269 configuration applied (or with
overrides merged on top of it).
Notes
Final-stage output is 2048 channels. Training such a deep network
benefits significantly from zero_init_residual=True and a long
warm-up schedule.
Examples
>>> import lucid
>>> from lucid.models.vision.resnet import resnet_269
>>> model = resnet_269()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape # (B, 2048, H/32, W/32)
(1, 2048, 7, 7)