Update nvsegment-ct readme

[heyufan1995]

No description provided.

[greptile-apps]

Greptile Summary

This PR expands and clarifies the NV-Segment-CT README and adds a minor wording improvement to the NV-Segment-CTMR README to make model-weight caching behavior explicit.

• NV-Segment-CT/docs/README.md: Fixes a broken GitHub link for label_dict.json (was pointing into NV-Segment-CTMR/configs/; now correctly points to NV-Segment-CT/configs/), and restructures the batch-inference section into distinct sub-sections (single-image, specific-class, single-GPU batch, multi-GPU batch) with detailed notes on resume semantics, discovery filters, input-list caching, and multi-GPU constraints. All referenced config keys and files are present in the repository.
• NV-Segment-CTMR/docs/README.md: One-line clarification appending "later runs reuse the cached weights." to the model-weights setup sentence.

Confidence Score: 5/5

Documentation-only changes; no runtime code is modified.

Both files are pure Markdown. The key factual fix — the GitHub URL for label_dict.json now correctly targets NV-Segment-CT/configs/ — was verified against the repository tree. All config keys (batch_resume_skip_existing, batch_use_input_list_cache, etc.) were confirmed in configs/batch_inference.json, and all referenced config files (inference.json, batch_inference.json, mgpu_inference.json) exist. Relative link ../configs/label_dict.json from NV-Segment-CT/docs/README.md resolves correctly.

No files require special attention.

Important Files Changed

Filename	Overview
NV-Segment-CT/docs/README.md	Major documentation expansion: fixes the GitHub link to label_dict.json (NV-Segment-CTMR → NV-Segment-CT), splits batch inference into dedicated sub-sections with detailed notes on resume, discovery filters, cache, and multi-GPU. All referenced config files and keys are verified to exist.
NV-Segment-CTMR/docs/README.md	Minor clarification: appends "later runs reuse the cached weights." to the model-weights sentence, making the caching behavior explicit. No functional changes.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User runs inference command] --> B{First run?}
    B -- Yes --> C[Download checkpoint from Hugging Face]
    C --> D[Link to models/model.pt]
    D --> E[Run inference]
    B -- No --> E[Run inference / reuse cached weights]

    E --> F{Batch or single image?}
    F -- Single image --> G[Save to output_dir/basename/basename_trans.nii.gz]
    F -- Batch --> H{batch_resume_skip_existing?}
    H -- true --> I[Queue only missing/empty outputs]
    H -- false --> J[Queue all discovered *.nii.gz]
    I --> K{Multi-GPU torchrun?}
    J --> K
    K -- Yes --> L[Rank 0 builds input_list, writes JSON cache]
    L --> M[Other ranks read cache]
    M --> N[Distribute volumes across GPUs]
    K -- No --> O[Single process runs all volumes]
    N --> G
    O --> G

Loading

_{Reviews (1): Last reviewed commit: "Update nvsegment-ct readme" | Re-trigger Greptile}