CHANGELOG.md

Note: This only highlights "important" changes. For more details, see commit history.

v0.3.11

• [Feature] Users may set ++misc.compile=True with PyTorch 2 to speed up training by compiling the model.
• [Feature] RD plot: allow users to specify results/**/*.json paths directly.
• [Feature] Simplify compressai-eval usage.
  To evaluate multiple models trained using CompressAI Trainer:

  compressai-eval \
      --config-path="$HOME/data/runs/e4e6d4d5e5c59c69f3bd7be2/configs" \
      --config-path="$HOME/data/runs/d4d5e5c5e4e6bd7be29c69f3/configs" \
      ...

  To evaluate multiple models from the CompressAI zoo:

  compressai-eval \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=1 ++criterion.lmbda=0.0018 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=2 ++criterion.lmbda=0.0035 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=3 ++criterion.lmbda=0.0067 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=4 ++criterion.lmbda=0.0130 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=5 ++criterion.lmbda=0.0250 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=6 ++criterion.lmbda=0.0483 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=7 ++criterion.lmbda=0.0932 \
      --config-name="eval_zoo" ++model.name="bmshj2018-factorized" ++model.quality=8 ++criterion.lmbda=0.1800

v0.3.10

• [Feature] compressai-plot --optimal=none|pareto|convex to select which points to display on RD curve.
• [Feature] Log git repository versions in new format, e.g. v0.3.9-8-g643ce8b-dirty.
• [Feature] GVAEImageCompressionRunner.

v0.3.9

• [Feature] CLI utilities now launch from:
  • compressai-train
  • compressai-eval (evaluate a trained model to produce bitstreams/images/metrics/etc)
  • compressai-plot (plot RD curves)
  • compressai_trainer.run.compressai (wrapper around compressai.utils)
• [Feature] Log x_hat images and debug_outputs to experiment tracker, too. (See v0.3.4 notes.)
• [CI] Automated tests.
• [Refactor] Simplify ImageCompressionRunner; expose Hydra configuration ++runner.meters and ++runner.inference (skip_compress/skip_decompress).

v0.3.8

• [Feature] Plot EntropyBottleneck distributions.
• [Fix] Ensure update(force=True) is called every epoch to update CDFs.

v0.3.7

• [Feature] CLI utilities reorganized into compressai_train.run.*
• [Refactor] ImageCompressionRunner.predict_batch now only predicts batches.
• [Chore] Upgrade various dependencies (poetry update).

v0.3.6

• [Feature] CLIC 2020 datasets.
• [Chore] Upgrade to aim==3.16.0.

v0.3.5

• [Feature] RD curves for MS-SSIM.
• [Refactor] Simplify ImageCompressionRunner by extracting loggers/etc.
• [Fix] CompressAI adapter for "psnr" being renamed to "psnr-rgb".

v0.3.4

• [Feature] Users of ImageCompressionRunner may now plot featuremaps/images during training by specifying "debug_outputs" in the forward/compress/decompress return dicts. For instance,

  def forward(self, x):
      ...
      return {
          "likelihoods": {"y": y_likelihoods},
          "debug_outputs": {
              "y_hat": y_hat,
              "means_hat": means_hat,
              "scales_hat": scales_hat,
              "nll": -y_likelihoods.log2(),
          },
      }

  The featuremaps are outputted in the configured paths.images directory.

v0.3.3

• [Chore] Update license copyright year to 2023.

v0.3.2

• [Feat] RD plot standard codecs (e.g. VTM).
• [Feat] Save runs/$RUN_HASH/configs/config.yaml.
• [Fix] Tensorboard logging.

v0.3.0

• [Refactor] Rename compressai_train -> compressai_trainer.

v0.2.17

• [Docs] Documentation! https://interdigitalinc.github.io/CompressAI-Trainer/

v0.2.16

• [Feat] Separately configurable optimizers for net/aux (e.g. Adam, etc).
• [Chore] Revert to aim==3.14.4.

v0.2.15

• [Feat] RD plot individual per-sample points. (e.g. kodim01, kodim02, ..., kodim24)
• [Chore] Upgrade to aim==3.15.1.