diff --git a/entcl/data/cifar100.py b/entcl/data/cifar100.py
index 5fbbf9df6df141b67f376ca197e298f999ca19c9..1e534d6783deef3d15d553e7932d252eadb6dbe4 100644
--- a/entcl/data/cifar100.py
+++ b/entcl/data/cifar100.py
@@ -159,9 +159,10 @@ class CIFAR100Dataset:
samples = torch.cat(samples)
labels = torch.cat(labels)
+ types = torch.full((labels.size(0),), 0, dtype=torch.long) # they are all known classes, so type is 0
logger.debug(f"Creating dataset for session 0 (pretraining). There are {len(samples)} samples, and {len(labels)} labels. There are {labels.unique().size(0)} different classes")
logger.debug(f"Classes in Pretraining Dataset: {labels.unique(sorted=True)}")
- datasets[0] = TransformedTensorDataset(tensor_dataset=torch.utils.data.TensorDataset(samples, labels), transform=self.transform)
+ datasets[0] = TransformedTensorDataset(tensor_dataset=torch.utils.data.TensorDataset(samples, labels, types), transform=self.transform)
# CL sessions' datasets
logger.debug("Splitting data for CL sessions")
diff --git a/entcl/run.py b/entcl/run.py
index 0fd351457fe7b9834c4b0a5e717c452bc0b790dc..4133e1ea018e9374330857c9625d97b82fe52a04 100644
--- a/entcl/run.py
+++ b/entcl/run.py
@@ -20,6 +20,8 @@ def main(args: argparse.Namespace):
logger.debug(f"Model: {model}")
logger.info("Pretraining Model (Session 0)")
+ args.current_session = 0
+ os.makedirs(os.path.join(args.exp_dir, 'session_0'), exist_ok=True)
model = pretrain(args, model)
if args.mode in ['cl', 'both']:
@@ -28,6 +30,8 @@ def main(args: argparse.Namespace):
for session in range(1, args.sessions + 1):
logger.info(f"Starting Continual Learning Session {session}")
args.current_session = session
+ os.makedirs(os.path.join(args.exp_dir, f'session_{session}'), exist_ok=True)
+
session_dataset = args.dataset.get_dataset(session)
# OOD detection
@@ -40,6 +44,7 @@ def main(args: argparse.Namespace):
# Expand Classification Head & Initialise
model.head.expand(args.dataset.novel_inc) # we are cheating here, we know the number of novel classes
+ assert model.head.fc.out_features == args.dataset.known + session * args.dataset.novel_inc, f"Head has {model.head.fc.out_features} features, expected {args.dataset.known + session * args.dataset.novel_inc}"
# freeze the weights for the existing classes. We are only training unknown samples (EG: 50 (known) + (2 (session) - 1) * 10 (novel_inc) = 60 classes have been seen in cl session 2)
model.head.freeze(start_idx=0, end_idx=args.dataset.known + ((session -1) * args.dataset.novel_inc))
@@ -150,14 +155,14 @@ if __name__ == "__main__":
if args.head == 'linear':
from entcl.models.linear_head import LinearHead
- args.head = LinearHead(in_features=768, out_features=args.dataset.num_classes)
+ args.head = LinearHead(in_features=768, out_features=args.dataset.known)
logger.debug(f"Using Linear Head: {args.head}")
elif args.head == 'dino_head':
from entcl.models.dinohead import DINOHead
args.head = DINOHead(768, args.dataset.known, nlayers=3)
elif args.head == 'mlp':
from entcl.models.linear_head import MLPHead
- args.head = MLPHead(in_features=768, out_features=args.dataset.num_classes, hidden_dim1=512, hidden_dim2=256)
+ args.head = MLPHead(in_features=768, out_features=args.dataset.known, hidden_dim1=512, hidden_dim2=256)
logger.debug(f"Using MLP Head: {args.head}")
if args.mode == 'cl' and args.pretrain_load is None:
diff --git a/enviroment.yml b/enviroment.yml
index 86f93904f4232f6a75749b06811cd61efbec2b4d..abc7d80627f122324dff6a5fa217e20b138b758d 100644
--- a/enviroment.yml
+++ b/enviroment.yml
@@ -6,16 +6,21 @@ channels:
- conda-forge
- defaults
dependencies:
- - _libgcc_mutex=0.1=main
- - _openmp_mutex=5.1=1_gnu
+ - _libgcc_mutex=0.1=conda_forge
+ - _openmp_mutex=4.5=2_gnu
- antlr-python-runtime=4.9.3=pyhd8ed1ab_1
- blas=1.0=mkl
+ - bottleneck=1.4.2=py310ha9d4c09_0
+ - brotli=1.0.9=h5eee18b_8
+ - brotli-bin=1.0.9=h5eee18b_8
- brotli-python=1.0.9=py310h6a678d5_8
- bzip2=1.0.8=h5eee18b_6
+ - c-ares=1.19.1=h5eee18b_0
- ca-certificates=2024.9.24=h06a4308_0
- certifi=2024.8.30=py310h06a4308_0
- cffi=1.17.1=py310h1fdaa30_0
- charset-normalizer=3.3.2=pyhd3eb1b0_0
+ - contourpy=1.2.0=py310hdb19cb5_0
- cpython=3.10.15=py310hd8ed1ab_2
- cuda-cudart=11.7.99=0
- cuda-cupti=11.7.101=0
@@ -24,49 +29,86 @@ dependencies:
- cuda-nvtx=11.7.91=0
- cuda-runtime=11.7.1=0
- cudatoolkit=11.3.1=h2bc3f7f_2
+ - cycler=0.11.0=pyhd3eb1b0_0
+ - cyrus-sasl=2.1.28=h52b45da_1
- dataclasses=0.8=pyh6d0b6a4_7
+ - dbus=1.13.18=hb2f20db_0
+ - expat=2.6.3=h6a678d5_0
- ffmpeg=4.3=hf484d3e_0
- filelock=3.13.1=py310h06a4308_0
+ - fontconfig=2.14.1=h55d465d_3
+ - fonttools=4.51.0=py310h5eee18b_0
- freetype=2.12.1=h4a9f257_0
- future=1.0.0=py310h06a4308_0
- fvcore=0.1.5.post20221221=pyhd8ed1ab_0
+ - glib=2.78.4=h6a678d5_0
+ - glib-tools=2.78.4=h6a678d5_0
- gmp=6.2.1=h295c915_3
- gmpy2=2.1.2=py310heeb90bb_0
- gnutls=3.6.15=he1e5248_0
+ - gst-plugins-base=1.14.1=h6a678d5_1
+ - gstreamer=1.14.1=h5eee18b_1
+ - icu=73.1=h6a678d5_0
- idna=3.7=py310h06a4308_0
- intel-openmp=2021.4.0=h06a4308_3561
- iopath=0.1.10=pyhd8ed1ab_0
- jinja2=3.1.4=py310h06a4308_1
+ - joblib=1.4.2=py310h06a4308_0
- jpeg=9e=h5eee18b_3
+ - kiwisolver=1.4.4=py310h6a678d5_0
+ - kneed=0.8.5=pyhd8ed1ab_0
+ - krb5=1.20.1=h143b758_1
- lame=3.100=h7b6447c_0
- lcms2=2.12=h3be6417_0
- ld_impl_linux-64=2.40=h12ee557_0
- lerc=3.0=h295c915_0
+ - libbrotlicommon=1.0.9=h5eee18b_8
+ - libbrotlidec=1.0.9=h5eee18b_8
+ - libbrotlienc=1.0.9=h5eee18b_8
+ - libclang=14.0.6=default_hc6dbbc7_1
+ - libclang13=14.0.6=default_he11475f_1
- libcublas=11.10.3.66=0
- libcufft=10.7.2.124=h4fbf590_0
- libcufile=1.9.1.3=0
+ - libcups=2.4.2=h2d74bed_1
- libcurand=10.3.5.147=0
+ - libcurl=8.9.1=h251f7ec_0
- libcusolver=11.4.0.1=0
- libcusparse=11.7.4.91=0
- libdeflate=1.17=h5eee18b_1
+ - libedit=3.1.20230828=h5eee18b_0
+ - libev=4.33=h7f8727e_1
- libffi=3.4.4=h6a678d5_1
- - libgcc-ng=11.2.0=h1234567_1
- - libgomp=11.2.0=h1234567_1
+ - libgcc=14.2.0=h77fa898_1
+ - libgcc-ng=14.2.0=h69a702a_1
+ - libgfortran-ng=11.2.0=h00389a5_1
+ - libgfortran5=11.2.0=h1234567_1
+ - libglib=2.78.4=hdc74915_0
+ - libgomp=14.2.0=h77fa898_1
- libiconv=1.16=h5eee18b_3
- libidn2=2.3.4=h5eee18b_0
+ - libllvm14=14.0.6=hecde1de_4
+ - libnghttp2=1.57.0=h2d74bed_0
- libnpp=11.7.4.75=0
- libnvjpeg=11.8.0.2=0
- libpng=1.6.39=h5eee18b_0
+ - libpq=17.0=hdbd6064_0
- libprotobuf=3.20.3=he621ea3_0
+ - libssh2=1.11.0=h251f7ec_0
- libstdcxx-ng=11.2.0=h1234567_1
- libtasn1=4.19.0=h5eee18b_0
- libtiff=4.5.1=h6a678d5_0
- libunistring=0.9.10=h27cfd23_0
- libuuid=1.41.5=h5eee18b_0
- libwebp-base=1.3.2=h5eee18b_1
+ - libxcb=1.15=h7f8727e_0
+ - libxkbcommon=1.0.1=h097e994_2
+ - libxml2=2.13.1=hfdd30dd_2
- loguru=0.7.2=py310h06a4308_1
- lz4-c=1.9.4=h6a678d5_1
- markupsafe=2.1.3=py310h5eee18b_0
+ - matplotlib=3.9.2=py310h06a4308_0
+ - matplotlib-base=3.9.2=py310hbfdbfaf_0
- mkl=2021.4.0=h06a4308_640
- mkl-service=2.4.0=py310h7f8727e_0
- mkl_fft=1.3.1=py310hd6ae3a3_0
@@ -74,43 +116,64 @@ dependencies:
- mpc=1.1.0=h10f8cd9_1
- mpfr=4.0.2=hb69a4c5_1
- mpmath=1.3.0=py310h06a4308_0
+ - mysql=8.4.0=h0bac5ae_0
- ncurses=6.4=h6a678d5_0
- nettle=3.7.3=hbbd107a_1
- networkx=3.3=py310h06a4308_0
- ninja=1.12.1=h06a4308_0
- ninja-base=1.12.1=hdb19cb5_0
+ - numexpr=2.8.4=py310h8879344_0
- numpy=1.24.3=py310hd5efca6_0
- numpy-base=1.24.3=py310h8e6c178_0
- omegaconf=2.3.0=pyhd8ed1ab_0
- openh264=2.1.1=h4ff587b_0
- openjpeg=2.5.2=he7f1fd0_0
- - openssl=3.0.15=h5eee18b_0
+ - openldap=2.6.4=h42fbc30_0
+ - openssl=3.4.0=hb9d3cd8_0
+ - pandas=2.2.2=py310h6a678d5_0
+ - pcre2=10.42=hebb0a14_1
- pillow=10.4.0=py310h5eee18b_0
- pip=24.2=py310h06a4308_0
+ - platformdirs=3.10.0=py310h06a4308_0
+ - ply=3.11=py310h06a4308_0
+ - pooch=1.8.2=py310h06a4308_0
- portalocker=2.3.0=py310h06a4308_1
- pycparser=2.21=pyhd3eb1b0_0
+ - pyparsing=3.2.0=py310h06a4308_0
+ - pyqt=5.15.10=py310h6a678d5_0
+ - pyqt5-sip=12.13.0=py310h5eee18b_0
- pysocks=1.7.1=py310h06a4308_0
- python=3.10.15=he870216_1
+ - python-dateutil=2.9.0post0=py310h06a4308_2
+ - python-tzdata=2023.3=pyhd3eb1b0_0
- python_abi=3.10=2_cp310
- pytorch=2.0.0=py3.10_cuda11.7_cudnn8.5.0_0
- pytorch-cuda=11.7=h778d358_5
- pytorch-mutex=1.0=cuda
- pyyaml=6.0.2=py310h5eee18b_0
+ - qt-main=5.15.2=hb6262e9_11
- readline=8.2=h5eee18b_0
- requests=2.32.3=py310h06a4308_1
+ - scikit-learn=1.5.1=py310h1128e8f_0
+ - seaborn=0.13.2=py310h06a4308_0
- setuptools=75.1.0=py310h06a4308_0
+ - sip=6.7.12=py310h6a678d5_0
- six=1.16.0=pyhd3eb1b0_1
- sqlite=3.45.3=h5eee18b_0
- sympy=1.13.3=pyh2585a3b_104
- tabulate=0.9.0=py310h06a4308_0
- termcolor=2.1.0=py310h06a4308_0
+ - threadpoolctl=3.5.0=py310h2f386ee_0
- tk=8.6.14=h39e8969_0
+ - tomli=2.0.1=py310h06a4308_0
- torchaudio=2.0.0=py310_cu117
- torchtriton=2.0.0=py310
- torchvision=0.15.0=py310_cu117
+ - tornado=6.4.1=py310h5eee18b_0
- tqdm=4.66.5=py310h2f386ee_0
- typing-extensions=4.11.0=py310h06a4308_0
- typing_extensions=4.11.0=py310h06a4308_0
+ - unicodedata2=15.1.0=py310h5eee18b_0
- urllib3=2.2.3=py310h06a4308_0
- wheel=0.44.0=py310h06a4308_0
- xformers=0.0.19=py310_cu11.8.0_pyt2.0.0
@@ -137,8 +200,8 @@ dependencies:
- distributed-ucxx-cu11==0.40.0
- fastrlock==0.8.2
- fsspec==2024.10.0
+ - gapstatistics==0.1.5
- importlib-metadata==8.5.0
- - joblib==1.4.2
- libcudf-cu11==24.10.1
- libucx-cu11==1.17.0
- libucxx-cu11==0.40.0
@@ -150,7 +213,6 @@ dependencies:
- numba==0.60.0
- nvtx==0.2.10
- packaging==24.2
- - pandas==2.2.2
- partd==1.4.2
- psutil==6.1.0
- ptxcompiler-cu11==0.8.1.post2
@@ -159,7 +221,6 @@ dependencies:
- pylibcudf-cu11==24.10.1
- pylibraft-cu11==24.10.0
- pynvml==11.4.1
- - python-dateutil==2.9.0.post0
- pytz==2024.2
- raft-dask-cu11==24.10.0
- rapids-dask-dependency==24.10.0
@@ -170,7 +231,6 @@ dependencies:
- submitit==1.5.2
- tblib==3.0.0
- toolz==1.0.0
- - tornado==6.4.1
- treelite==4.3.0
- tzdata==2024.2
- ucx-py-cu11==0.40.0