Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.
日本語の概要は準備中です。原文の説明を表示しています。
Routes DI-engine users to the right workflow for CLI/config launches, serial and framework-based RL pipelines, and environment integration.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Use this skill for the public DI-engine reinforcement-learning engine. It routes common user requests to focused sub-skills instead of forcing future agents to rediscover the repo structure from scratch.
If you only need to verify that the package is installed, start with:
python -m pip check
python -c "import ding, dizoo; print(ding.__version__)"
For a faster bundled smoke check, run scripts/check_install.py from an
environment where DI-engine is installed.
cli-configRead this when the user asks about:
ding or ditask CLI usagemain_config, create_config, or system_configcompile_config, read_config, save_config, or registry queriesserial, parallel, dist, eval, or ditaskserial-pipelinesRead this when the user asks about:
ding.entry.serial_pipeline, serial_pipeline_onpolicy, or serial_pipeline_offlineeval, collect_demo_data, collect_episodic_demo_data, or episodic post-processingding/entry/* launchers and environment-specific entry scripts under
dizoo/framework-runtimeRead this when the user asks about:
ding.framework.task, Parallel, Supervisor, or EventLoopStepCollector, OffPolicyLearner,
multistep_trainer, interaction_evaluator, CkptSaver, or data_pusherding/example/task.wait_for, task.emit, or auto-recoveryenv-integrationRead this when the user asks about:
DingEnvWrapper, BaseEnvManager, BaseEnvManagerV2, or EnvSupervisordizoo/classic_control/, bitflip/,
frozen_lake/, or league_demo/scripts/check_install.py — quick import/version/config smoke for a prepared environment.scripts/run_serial.sh — thin wrapper for ding -m serial with a config path and seed.scripts/run_parallel.sh — thin wrapper for ding -m parallel with a config path and seed.references/package-surface.md for the repo module map and public entry points.references/troubleshooting.md for cross-cutting install/import/runtime failures.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.
日本語の概要は準備中です。原文の説明を表示しています。
Guide 3DDFA Python inference, geometry rendering, training/evaluation, and optional C++ ONNX workflows for 3D dense face alignment.
日本語の概要は準備中です。原文の説明を表示しています。
Routes 3DDFA_V2 face-alignment setup, still-image demos, video tracking, and ONNX benchmarking workflows.
日本語の概要は準備中です。原文の説明を表示しています。
Operate AB3DMOT 3D multi-object tracking workflows for KITTI and nuScenes data, tracking, evaluation, and visualization.
日本語の概要は準備中です。原文の説明を表示しています。
Use Hugging Face Accelerate for PyTorch training-loop migration, distributed launch/configuration, DeepSpeed/FSDP/TPU backend setup, big-model inference/offload, checkpointing, tracking, and troubleshooting.
日本語の概要は準備中です。原文の説明を表示しています。
Route Acme reinforcement-learning framework tasks across core loops, replay/data, JAX agents, and TensorFlow/Sonnet agents.
日本語の概要は準備中です。原文の説明を表示しています。