Thanks for your great work.

I ran your 'contrastive_da.sh' file with

DATA=cifar10ood SEED=3 CATEGORY=0

But I got

{ "embed.auc": 63.617005555555565, "embed.gde": 61.748266666666666, "embed.kde": 64.29441111111112, "embed.kocsvm": 63.649888888888896, "embed.locsvm": 66.37852222222223, "pool.auc": 77.99122777777778, "pool.gde": 80.53367777777778, "pool.kde": 74.89410000000001, "pool.kocsvm": 75.82452222222223, "pool.locsvm": 67.06372222222222 }.

And with other seed 72

I got

{ "embed.auc": 50.0, "embed.gde": 49.977777777777774, "embed.kde": 50.0, "embed.kocsvm": 49.977777777777774, "embed.locsvm": 50.022222222222226, "pool.auc": 60.998122222222214, "pool.gde": 60.27448888888889, "pool.kde": 56.09522222222223, "pool.kocsvm": 55.180677777777774, "pool.locsvm": 51.08432222222222 }

Could you let me know how to reproduce your results?! Thank you:)

Thanks for your great work. I notice that you are also the authors of paper "CutPaste: Self-Supervised Learning for Anomaly Detection and Localization", and I can't reproduce the Cutpaste(normal cutpaste) results of Experiment with ResNet18 in A.1 section.
About the experiment setting: Backbone: Resnet 18 + MLP head(Train from scratch) For example, Capsule in mvtec dataset, training epoches: 256, num of training samples: 219, batchsize:64, one epoch needs 4 steps. In paper, "Note that, unlike conventional definition for an epoch, we define 256 parameter update steps as one epoch." So 65536 steps. the other parameters i set is lined with paper(including learning rate, weight_decay, moment). The training loss curve:

The acc curve:

The lr curve:

The epoch curve:

I'm not sure 65536 steps is too many. but according the loss curve, it' kind of weird. Finally the ROCAUC is , (paper: 87.9+-0.7), I think my evaluating is correct. I try evaluating pretrained efficientnetb4 and b5 without training(paper, table 3)

So about the experiment of resnet18, Could you please give me some advice?

If I understand the code correctly, then the experiments with projection heads (e.g. script/run_contrastive_da.sh) use the projection head at test time and not just training time. This is from looking at resnet_util.ResNet which looks like the output embeds is always the projection head. So, in the notation of the paper (e.g. figure 4), the contrastive results generated by this repo are all Contrastive (DA) g o f

So then in table 2 of the paper where it says Contrastive (DA) that refers to using g o f`` since the scriptscript/run_contrastive_da.sh` is supposed to reproduce that row?

That's a little confusing to me since table 4 shows that the best results are obtained without using the projection head at test time (i.e. in the second stage).

Is my understanding correct?

Hi, I am currently having problem trying to replicate the results from celeba. I am running the run_contrastive.sh with these Settings:

DATA=celeba
METHOD=Contrastive
SEED=1
CATEGORY=Eyeglasses
MODEL_DIR='.'
python train_and_eval_loop.py \
  --model_dir="${MODEL_DIR}" \
  --method=${METHOD} \
  --file_path="${DATA}_${PREFIX}_s${SEED}_c${CATEGORY}" \
  --dataset=${DATA} \
  --category=${CATEGORY} \
  --seed=${SEED} \
  --root='' \
  --net_type=ResNet18 \
  --net_width=1 \
  --latent_dim=0 \
  --aug_list="cnr0.5+hflip+jitter_b0.4_c0.4_s0.4_h0.4+gray0.2+blur0.5,+" \
  --aug_list_for_test="x" \
  --input_shape="64,64,3" \
  --optim_type=sgd \
  --sched_type=cos \
  --learning_rate=0.01 \
  --momentum=0.9 \
  --weight_decay=0.0003 \
  --head_dims="512,512,512,512,512,512,512,512,128" \
  --num_epoch=2048 \
  --batch_size=32 \
  --temperature=0.2 \
  --distaug_type 1

the error I get is this:

/test/contrastive.py:87 step_fn  *
        y = self.get_target_labels(
    /test/contrastive.py:69 get_target_labels  *
        x_concat = self.cross_replica_concat(x, replica_context=replica_context)
    /test/util/train.py:792 cross_replica_concat  *
        ext_tensor = tf.scatter_nd(
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/gen_array_ops.py:8856 scatter_nd  **
        "ScatterNd", indices=indices, updates=updates, shape=shape, name=name)
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:479 _apply_op_helper
        repr(values), type(values).__name__, err))

    TypeError: Expected int32 passed to parameter 'shape' of op 'ScatterNd', got [3, None, 64, 64, 3] of type 'list' instead. Error: Expected int32, got None of type 'NoneType' instead.

can you help me fix this?