prediction-and-inference
Routes EasyCV predictor APIs, batch prediction, feature extraction, and exported-model inference.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- prediction-and-inference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Routes EasyCV predictor APIs, batch prediction, feature extraction, and exported-model inference.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/batch-prediction.md1.8 KB
- 参考文档references/predictors.md2.9 KB
- 参考文档references/troubleshooting.md1.7 KB
03 SKILL.md 正文
Prediction and inference
Use this sub-skill when the task is to run EasyCV inference from Python, a batch file or ODPS table, or an exported raw / JIT / Blade / ONNX-style model.
It covers the predictor family under easycv.predictors and the batch front door python -m easycv.tools.predict.
Read these references first
references/predictors.mdfor the major predictor classes and their shared constructor patterns.references/batch-prediction.mdfor file-list and table-based batch inference.references/troubleshooting.mdfor missing extras, model-type mismatches, and shape / label issues.- Root
references/cli-reference.mdfor the installed-package command front doors.
What belongs here
Include tasks such as:
- running classification, detection, segmentation, pose, OCR, video, ReID, feature-extraction, or BEVFormer inference
- selecting the right predictor class for the exported artifact you already have
- loading raw, JIT, Blade, or ONNX-style model files
- batch prediction over local files, URLs, or ODPS tables
- label-map handling, score thresholds, and
topkoutput shaping - image-mode issues such as
BGRvsRGB
What stays elsewhere
- Training / validation / config selection ->
sub-skills/training-and-evaluation/ - Export and optimization of the model artifact ->
sub-skills/export-and-optimization/ - Dataset conversion, file layouts, and OSS setup ->
sub-skills/data-preparation/
Typical decision flow
- Decide whether you want a direct predictor or the batch tool.
- Identify the model family and the artifact format.
- Check whether you have a config file, label map, or exported sidecar files.
- Confirm the input form: file path, image array, list file, or ODPS table.
- Run the smallest safe inference path first, then scale to batches or tables.
Common success signals
- The predictor class matches the model artifact type.
- The input processor accepts the image mode and file format.
- The output processor returns the expected
class,detection_boxes,keypoints,feature, or similar fields. - Batch inference writes the expected output file or table columns.
Common prediction surfaces
easy_predictis required for the batch-file / table CLI.- Exported JIT / Blade models usually need their sidecar config or preprocess artifact nearby.
mode='RGB'is common for PIL-based classification; many detection pipelines remainBGR.label_map_pathorCLASSEScontrols readable class names.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.