fix v0.2.0

README.md

@@ -24,6 +24,7 @@ This repo will be continued to update in the next two months! Please watch us fo
 
 Please refer to [CHANGELOG.md](docs/CHANGELOG.md) for details and release history.
 
+[2020-04-08] Configs reoriganized and new methods supported in `OpenMixup` v0.2.0.
 [2020-03-31] `OpenMixup` v0.2.0 is released.
 
 ## Installation
@@ -37,7 +38,7 @@ Then, see [tutorials](docs/tutorials) for more tech details (based on MMClassifi
 
 ## Benchmark and Model Zoo
 
-[Model Zoos](docs/model_zoos) and list of [Awesome Mixups](docs/awesome_mixups) will be updated in the next two months (before 2022.04)!
+[Model Zoos](docs/model_zoos) and lists of [Awesome Mixups](docs/awesome_mixups) have been released, and will be updated in the next two months. Checkpoints and traning logs will be updated soon! 
 
 ## License
 
@@ -50,12 +51,12 @@ This project is released under the [Apache 2.0 license](LICENSE).
 
 ## Citation
 
-If you find this project useful in your research, please consider cite:
+If you find this project useful in your research, please consider cite our repo:
 
 ```BibTeX
 @misc{2022openmixup,
-    title={{OpenMixup}: Open Mixup Toolbox and Benchmark for Visual Representation},
-    author={Li, Siyuan and Liu, Zichen and Wu, Di},
+    title={{OpenMixup}: Open Mixup Toolbox and Benchmark for Visual Representation Learning},
+    author={Li, Siyuan and Liu, Zichen and Wu, Di, Stan Z. Li},
     howpublished = {\url{https://github.com/Westlake-AI/openmixup}},
     year={2022}
 }

benchmarks/classification/knn_imagenet/dist_test_knn_epoch.sh

@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+
+set -e
+set -x
+
+CFG=$1
+EPOCH=$2
+PY_ARGS=${@:3}
+GPUS=${GPUS:-8}
+NNODES=${NNODES:-1}
+NODE_RANK=${NODE_RANK:-0}
+PORT=${PORT:-29500}
+MASTER_ADDR=${MASTER_ADDR:-"127.0.0.1"}
+
+WORK_DIR=$(echo ${CFG%.*} | sed -e "s/configs/work_dirs/g")/
+
+if [ ! -f $WORK_DIR/epoch_${EPOCH}.pth ]; then
+    echo "ERROR: File not exist: $WORK_DIR/epoch_${EPOCH}.pth"
+    exit
+fi
+
+python -m torch.distributed.launch \
+    --nnodes=$NNODES \
+    --node_rank=$NODE_RANK \
+    --master_addr=$MASTER_ADDR \
+    --nproc_per_node=$GPUS \
+    --master_port=$PORT \
+    tools/benchmarks/classification/knn_imagenet/test_knn.py $CFG \
+    --checkpoint $WORK_DIR/epoch_${EPOCH}.pth \
+    --work-dir $WORK_DIR --launcher="pytorch" ${PY_ARGS}

benchmarks/classification/knn_imagenet/dist_test_knn_pretrain.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+
+set -e
+set -x
+
+CFG=$1
+PRETRAIN=$2  # pretrained model
+PY_ARGS=${@:3}
+GPUS=${GPUS:-8}
+NNODES=${NNODES:-1}
+NODE_RANK=${NODE_RANK:-0}
+PORT=${PORT:-29500}
+MASTER_ADDR=${MASTER_ADDR:-"127.0.0.1"}
+
+# set work_dir according to config path and pretrained model to distinguish different models
+WORK_DIR="$(echo ${CFG%.*} | sed -e "s/configs/work_dirs/g")/$(echo $PRETRAIN | rev | cut -d/ -f 1 | rev)"
+
+python -m torch.distributed.launch \
+    --nnodes=$NNODES \
+    --node_rank=$NODE_RANK \
+    --master_addr=$MASTER_ADDR \
+    --nproc_per_node=$GPUS \
+    --master_port=$PORT \
+    tools/benchmarks/classification/knn_imagenet/test_knn.py $CFG \
+    --cfg-options model.backbone.init_cfg.type=Pretrained \
+    model.backbone.init_cfg.checkpoint=$PRETRAIN \
+    --work-dir $WORK_DIR --launcher="pytorch" ${PY_ARGS}

benchmarks/classification/knn_imagenet/knn_classifier.py

@@ -0,0 +1,70 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
+# This file is borrowed from
+# https://github.com/facebookresearch/dino/blob/main/eval_knn.py
+
+import torch
+import torch.nn as nn
+
+
+@torch.no_grad()
+def knn_classifier(train_features,
+                   train_labels,
+                   test_features,
+                   test_labels,
+                   k,
+                   T,
+                   num_classes=1000):
+    """Compute accuracy of knn classifier predictions.
+
+    Args:
+        train_features (Tensor): Extracted features in the training set.
+        train_labels (Tensor): Labels in the training set.
+        test_features (Tensor): Extracted features in the testing set.
+        test_labels (Tensor): Labels in the testing set.
+        k (int): Number of NN to use.
+        T (float): Temperature used in the voting coefficient.
+        num_classes (int): Number of classes. Defaults to 1000.
+    """
+    top1, top5, total = 0.0, 0.0, 0
+    train_features = nn.functional.normalize(train_features, dim=1)
+    test_features = nn.functional.normalize(test_features, dim=1)
+    train_features = train_features.t()
+    num_test_images, num_chunks = test_labels.shape[0], 100
+    # split all test images into several chunks to prevent out-of-memory
+    imgs_per_chunk = num_test_images // num_chunks
+    retrieval_one_hot = torch.zeros(k, num_classes).to(train_features.device)
+    for idx in range(0, num_test_images, imgs_per_chunk):
+        # get the features for test images
+        features = test_features[idx:min((idx +
+                                          imgs_per_chunk), num_test_images), :]
+        targets = test_labels[idx:min((idx + imgs_per_chunk), num_test_images)]
+        batch_size = targets.shape[0]
+
+        # calculate the dot product and compute top-k neighbors
+        similarity = torch.mm(features, train_features)
+        distances, indices = similarity.topk(k, largest=True, sorted=True)
+        candidates = train_labels.view(1, -1).expand(batch_size, -1)
+        retrieved_neighbors = torch.gather(candidates, 1, indices)
+
+        retrieval_one_hot.resize_(batch_size * k, num_classes).zero_()
+        retrieval_one_hot.scatter_(1, retrieved_neighbors.view(-1, 1), 1)
+        distances_transform = distances.clone().div_(T).exp_()
+        probs = torch.sum(
+            torch.mul(
+                retrieval_one_hot.view(batch_size, -1, num_classes),
+                distances_transform.view(batch_size, -1, 1),
+            ),
+            1,
+        )
+        _, predictions = probs.sort(1, True)
+
+        # find the predictions that match the target
+        correct = predictions.eq(targets.data.view(-1, 1))
+        top1 = top1 + correct.narrow(1, 0, 1).sum().item()
+        top5 = top5 + correct.narrow(1, 0, min(
+            5, k)).sum().item()  # top5 does not make sense if k < 5
+        total += targets.size(0)
+    top1 = top1 * 100.0 / total
+    top5 = top5 * 100.0 / total
+    return top1, top5

benchmarks/classification/knn_imagenet/slurm_test_knn_epoch.sh

@@ -0,0 +1,36 @@
+#!/usr/bin/env bash
+
+set -e
+set -x
+
+PARTITION=$1
+JOB_NAME=$2
+CFG=$3
+EPOCH=$4
+PY_ARGS=${@:5}
+GPUS=${GPUS:-8}
+GPUS_PER_NODE=${GPUS_PER_NODE:-8}
+CPUS_PER_TASK=${CPUS_PER_TASK:-5}
+PORT=${PORT:-29500}
+SRUN_ARGS=${SRUN_ARGS:-""}
+
+WORK_DIR=$(echo ${CFG%.*} | sed -e "s/configs/work_dirs/g")/
+
+if [ ! -f $WORK_DIR/epoch_${EPOCH}.pth ]; then
+    echo "ERROR: File not exist: $WORK_DIR/epoch_${EPOCH}.pth"
+    exit
+fi
+
+PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \
+srun -p ${PARTITION} \
+    --job-name=${JOB_NAME} \
+    --gres=gpu:${GPUS_PER_NODE} \
+    --ntasks=${GPUS} \
+    --ntasks-per-node=${GPUS_PER_NODE} \
+    --cpus-per-task=${CPUS_PER_TASK} \
+    --kill-on-bad-exit=1 \
+    ${SRUN_ARGS} \
+    python -u tools/benchmarks/classification/knn_imagenet/test_knn.py $CFG \
+        --checkpoint $WORK_DIR/epoch_${EPOCH}.pth \
+        --cfg-options dist_params.port=$PORT \
+        --work-dir $WORK_DIR --launcher="slurm" ${PY_ARGS}

benchmarks/classification/knn_imagenet/slurm_test_knn_pretrain.sh

@@ -0,0 +1,33 @@
+#!/usr/bin/env bash
+
+set -e
+set -x
+
+PARTITION=$1
+JOB_NAME=$2
+CFG=$3
+PRETRAIN=$4  # pretrained model
+PY_ARGS=${@:5}
+GPUS=${GPUS:-8}
+GPUS_PER_NODE=${GPUS_PER_NODE:-8}
+CPUS_PER_TASK=${CPUS_PER_TASK:-5}
+PORT=${PORT:-29500}
+SRUN_ARGS=${SRUN_ARGS:-""}
+
+# set work_dir according to config path and pretrained model to distinguish different models
+WORK_DIR="$(echo ${CFG%.*} | sed -e "s/configs/work_dirs/g")/$(echo $PRETRAIN | rev | cut -d/ -f 1 | rev)"
+
+PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \
+srun -p ${PARTITION} \
+    --job-name=${JOB_NAME} \
+    --gres=gpu:${GPUS_PER_NODE} \
+    --ntasks=${GPUS} \
+    --ntasks-per-node=${GPUS_PER_NODE} \
+    --cpus-per-task=${CPUS_PER_TASK} \
+    --kill-on-bad-exit=1 \
+    ${SRUN_ARGS} \
+    python -u tools/benchmarks/classification/knn_imagenet/test_knn.py $CFG \
+        --cfg-options model.backbone.init_cfg.type=Pretrained \
+        model.backbone.init_cfg.checkpoint=$PRETRAIN \
+        dist_params.port=$PORT \
+        --work-dir $WORK_DIR --launcher="slurm" ${PY_ARGS}