Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Hierarchical Pretraining: Research Repository

This is a research repository for reproducing the results from the project "Self-supervised pretraining improves self-supervised pretraining." You can find the arXiv prepint here:

  title={Self-supervised pretraining improves self-supervised pretraining.},
  author={Reed, Colorado J and Yue, Xiangyu and Nrusimha, Ani and Ebrahimi, Sayna and Vijaykumar, Vivek and Mao, Richard and Li, Bo and Zhang, Shanghang and Guillory, Devin and Metzger, Sean and Keutzer, Kurt and Darrell, Trevor},
  journal={arXiv preprint arXiv:2103.12718},

Pretrained Models

Note: we are currently adding many more pretrained models to this table.


Pretraining Algorithm Pretraining Data Pretraining Amount Type Notes Model URL
MoCo-v2 ImageNet 200 ep Pytorch 67.7% ImageNet Top-1. codebase - paper download
MoCo-v2-800 ImageNet 800 ep Pytorch 71.1% ImageNet Top-1. codebase - paper download
SimCLR-v1 ImageNet 800 ep Tensorflow 69.1% ImageNet Top-1. codebase - paper download
SimCLR-v2 ImageNet 800 ep Tensorflow 71.7% ImageNet Top-1. codebase - paper Google Storage (r50_1x_sk0/model.ckpt...)


# repo
git clone

# setup environment
conda create -n hpt python=3.7 ipython
conda activate hpt

# NOTE: if you are not using CUDA 10.2, you need to change the 10.2 in this file appropriately
# (check CUDA version with e.g. `cat /usr/local/cuda/version.txt`)
conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch

# install local submodules
cd OpenSelfSup
pip install -v -e .

Data installation

The data you need will depend on your goals, but generally speaking, download the RESISC-45 dataset to make sure you have everything working correctly.

Pretrained Models

cd OpenSelfSup/data/basetrain_chkpts/


RESISC-45 contains 31,500 aerial images, covering 45 scene classes with 700 images in each class.

# cd to the directory where you want the data, $DATA
tar xf resisc45.tar.gz

# replace/set $DATA and $CODE as appropriate
ln -s $DATA/resisc45 $CODE/OpenSelfSup/data/resisc45/all

Verify Install

Check installation by pretraining using mocov2, extracting the model weights, evaluating the representations, and then viewing the results:

cd OpenSelfSup

# Sanity check: MoCo for 20 epoch on 4 gpus
./tools/ configs/selfsup/moco/ 4

# make some variables so its clear what's happening
# Extract the backbone
python tools/ ${CHECKPOINT} ${BACKBONE}

# Evaluate the representations
./benchmarks/ configs/benchmarks/linear_classification/resisc45/ ${BACKBONE}

# View the results
cd work_dirs
# you may need to install tensorboard
tensorboard --logdir .

Base Training

Right now we assume ImageNet base trained models.

cd OpenSelfSup/data/basetrain_chkpts/

Pretraining With a New Dataset

We have a handy set of config generators to make pretraining with a new dataset easy and consistent!

FIRST, you will need the image pixel mean/std of your dataset, if you don't have it, you can do:

./ --data /path/to/image-folder --numworkers 20 --batchsize 256

where image-folder has the structure from ImageFolder in pytorch
or whatever image extension you're using

if your dataset is not arranged in this way, you can either: (i) use symlinks to put it in this structure (ii) update the above script to read in your data

NEXT, copy the pretraining template

cd utils
cp templates/ pretrain-configs/
# edit pretrain-configs/

# once edited, generate the project
./ pretrain-configs/

What just happened? We generated a bunch of pretraining configs in the following location (take a loot at all of these files to get a feel for how this works):


NEXT, you're ready to kick off a trial run to make sure the pretraining is working as expected =)

# the `-t` flag means `trial`: it'll only run 1 20 epoch pretraining
 ./utils/ -t -d OpenSelfSup/configs/hpt-pretrain/${shortname}

NEXT, if this works, kick off the full training. NOTE: you can kick this off multiple times as long as the config directories share the same filesystem

# simply removing the `-t` flag from above
 ./utils/ -d OpenSelfSup/configs/hpt-pretrain/${shortname}

NEXT, if you want to perform BYOL pretraining, add -b flag.

# simply add the `-b` flag to above. Currently, we only do it on Chexpert, Resisc, and Bdd for Exp3
 ./utils/ -d OpenSelfSup/configs/hpt-pretrain/${shortname} -b

Congratulations: you've launch a full hierarchical pretraining experiment.


  • How does keep track of what's been pretrained?
    • In each config directory, it creates a .pretrain-status folder to keep track of what's processing/finished. See them with e.g. find OpenSelfSup/configs/hpt-pretrain -name '.pretrain-status'
  • How to redo a pretraining, e.g. because it crashed or something changed? Remove the
    • Remove the associate .proc or .done file. Find these e.g.
    find OpenSelfSup/configs/hpt-pretrain -name '.proc'
    find OpenSelfSup/configs/hpt-pretrain -name '.done'

Evaluating Pretrained Representations

This has been simplified to simply:

./utils/ -b OpenSelfSup/work_dirs/hpt-pretrain/${shortname}/ -d OpenSelfSup/configs/hpt-pretrain/${shortname}

where -b is the backbone directory and -d is the config directory. This command also works for cross-dataset evaluation (e.g. evaluate models trained on Resic45 and evaluate on UC Merced dataset).


Where are the checkpoints and logs? E.g., if you pass in configs/hpt-pretrain/resisc as the config directory, then the working directories for this evalution is e.g. work_dirs/hpt-pretrain/resisc/linear-eval/....


Assuming you generated the pretraining project as specified above, finetuning is as simple as:

./utils/ -d ./OpenSelfSup/configs/hpt-pretrain/${shortname}/finetune/ -b ./OpenSelfSup/work_dirs/hpt-pretrain/${shortname}/

where -b is the backbone directory and -d is the config directory Note: to finetune using other backbones, simply pass in a different backbone directory (the script searches for final_backbone.pth files in the provided directory tree)

Finetuning only on pretrained checkpoints with BEST linear analysis

First, specify the pretraining epochs which gives the best linear evaluation result in ./utils/top-linear-analysis-ckpts.txt. Here is an example:

# dataset best-moco-bt best-sup-bt best-no-bt
chest_xray_kids 5000 10000 100000
resisc 5000 50000 100000
chexpert 50000 50000 400000

, in which for chest_xray_kids dataset, 5000-iters, 10000-iters, 100000-iters are the best pretrained models under moco base-training, imagenet-supervised base-training, and no base-training, respectively.

Second, run the following command to perform finetuning only on the best checkpoints (same as above, except that the change of script name):

./utils/ -d ./OpenSelfSup/configs/hpt-pretrain/${shortname}/finetune/ -b ./OpenSelfSup/work_dirs/hpt-pretrain/${shortname}

Pretraining on top of pretraining

Using the output of previously pretrained models, it is very easy to correctly setup pretraining on top of the pretraining. Simply create a new config


(see for an example)

and then set the basetrained models to be the final_backbone.pth from the output of the last pretrained. e.g. for using resisc-45 outputs:

export basetrain_weights=(



(see for an example)

To select which backbones to use, evaluate the linear performance from the various source outputs (e.g. all the resisc pretrained outputs) on the target data (e.g. on uc-merced data).

Then simply generate the project and execute the pretraining as normal:

./ pretrain-configs/

./ -d OpenSelfSup/configs/hpt-pretrain/$dataname1-dataname2

Object Detection / Semantic Segmentation

Object detection/segmentation uses detectron2 and takes place in the directory


First: Check if the dataset configs you need are already present in configs. E.g. if you're working with CoCo, you'll see the following 2 configs:


We'll use the config with the _moco suffix for all obj det and segmentation. If your configs already exist, skip the next step.

Next: assuming your configs do not exist, set up the configs you need for your dataset by copying an existing set of configs

cp configs/coco_R_50_C4_2x.yaml ${MYDATA}_R50_C4_2x.yaml
cp configs/coco_R_50_C4_2x_moco.yaml ${MYDATA}_R50_C4_2x_moco.yaml

Edit ${MYDATA}_R50_C4_2x.yaml and set MIN_SIZE_TRAIN and MIN_SIZE_TEST to be appropriate for your dataset. Also, rename TRAIN and TEST to have your dataset name, set MASK_ON to True if doing semantic segmentation, and update STEPS and MAX_ITER if running the training for a different amount of time is appropriate (check relevant publications / codebases to set the training schedule).

Edit ${MYDATA}_R50_C4_2x_moco.yaml and set PIXEL_MEAN and PIXEL_STD (use script above, if you don't know them).

Then, edit and add the appropriate data registry lines for your train/val data

register_coco_instances("dataname_train", {}, "obj-labels-in-coco-format_train.json", "datasets/dataname/dataname_train")
register_coco_instances("dataname_val", {}, "obj-labels-in-coco-format_val.json", "datasets/dataname/dataname_val")

Then, setup symlinks to your data under datasets/dataname/dataname_train and datasets/dataname/dataname_val, where you replace dataname with your dataname used in the config/registry.

Next, convert your backbone(s) to detectron format, e.g. (NOTE: I recommend keeping backbones in the same directory that they are originally present in, and appending a -detectron2 suffix)

python ../../data/basetrain_chkpts/imagenet_r50_supervised.pth ../../data/basetrain_chkpts/imagenet_r50_supervised-detectron2.pth

Next kick off training

python --config-file configs/DATANAME_R_50_C4_24k_moco.yaml --num-gpus 4 OUTPUT_DIR results/${UNIQUE_DATANAME_EXACTLY_DESCRIBING_THIS_RUN}/ TEST.EVAL_PERIOD 2000 MODEL.WEIGHTS ../../data/basetrain_chkpts/imagenet_r50_supervised-detectron2.pth SOLVER.CHECKPOINT_PERIOD ${INT_HOW_OFTEN_TO_CHECKPOINT}

results will be in results/${UNIQUE_DATANAME_EXACTLY_DESCRIBING_THIS_RUN}, and you can use tensorboard to view them.

Commit and Share Results

Run the following command to grab all results (linear analysis, finetunes, etc) and put them into the appropriate json results file in results/:


You can verify the results in results and then add the new/updated results file to git and commit.

Did you get an error message such as:


Please investigate as your results may not be complete.
(see errors in file: base-training/utils/tmp/errors.txt)

will not include partial result for base-training/utils/../OpenSelfSup/work_dirs/hpt-pretrain/resisc/finetune/1000-labels/imagenet_r50_supervised_basetrain/50000-iters-2500-iter-0_01-lr-finetune/20200911_170916.log.json

This means that this particular evaluation run did not appear to run for enough iterations. Investigate the provided log file, rerun any necessary evaluations, and remove the offending log file.

Debugging this script this script finds the top val accuracy, and save the corresponding test accuracy using the following script:


which outputs results to utils/tmp/results.txt and errors to utils/tmp/errors.txt. Look at this file if your results aren't being generated correctly.

Generate plots

cd utils

See plots in directory plot-results (you can also pass in a --data flag to only generate plots for a specific dataset, e.g. python --data resisc)

To plot the eval & test acc curves, use ./utils/

cd utils

To Generate plot for full finetuning, do

bash utils/

See plot in directory plot-results/exp-2.

To Generate plot for HPT Pretraining, do

bash utils/

See plot in directory plot-results/exp-3.

Getting activations for similarity measures

Run with a model used for a classifaction task (one that has a test/val set).
Alternatively, run dist_get_acts as follows:

bash ${CFG} ${CHECKPOINT} [--grab_conv...]

Default behavior is to grab the entire batch of linear layers. Setting --grab_conv will capture a single batch of all convolutional layers.
Layers will be saved in ${WORK_DIR}/model_acts.npz. The npz contains a dictionary which maps layer names to the activations. TODO: add similarity documentation. Similarity measures will be a seperate PR.

Important Notes:

  • If not using the provided experiment management scripts: add a new config file for each base training, even small variations, since the evaluation stage assumes a unique filename for each config (otherwise the checkpoints will overwrite the .previous versions).
  • If not using 4 GPUs, update the learning rate to new_lr = old_lr * new_ngpus / 4

Adding new repositories

  1. Add new repositories as a subtree to the base repository (next to this readme), e.g. this is what I used for OpenSelfSup (subtree cheatsheet):
git remote add openselfsup
git fetch openselfsup
git subtree add --prefix=OpenSelfSup openselfsup/master --squash
  1. After updating your conda env for the new repo, update the conda-env.yml file as follows:
conda env export --from-history > conda-env.yml

Debugging and Developing Within OpenSelfSup

Here's a command that will allow breakpoints (WARNING: the results with the debug=true flag SHOULD NOT BE USED -- they disable sync batch norms and are not comparable to other results):

# from OpenSelfSup/
# replace with your desired config
python tools/ configs/hpt-pretrain/resisc/moco_v2_800ep_basetrain/ --work_dir work_dirs/debug --debug


Repository providing a wide range of self-supervised pretrained models for computer vision tasks.






No releases published


No packages published