skills.rest
DeepLink-org
DeepLink-orgOFFICIAL
Explore all skills by this author →
WebsiteGitHub

nvidia-cv-segmentation

Benchmarks semantic segmentation training performance on NVIDIA GPUs using mmsegmentation.

7|1|Updated Apr 17, 2026
Search Tags:#benchmark#pytorch#performance evaluation#distributed training#nvidia#semantic segmentation#mmsegmentation
One-click install
npx skills add https://github.com/DeepLink-org/DeepEval-Skills --skill nvidia-cv-segmentation-deeplink-org
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: nvidia-cv-segmentation
Source: https://github.com/DeepLink-org/DeepEval-Skills/tree/main/skills/NVIDIA/cv/nvidia-cv-segmentation
Command: npx skills add https://github.com/DeepLink-org/DeepEval-Skills --skill nvidia-cv-segmentation-deeplink-org

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires torch, mmengine, mmcv, mmsegmentation, and includes scripts (resource) components.

What problem does it solve? Measuring training performance of semantic segmentation models on NVIDIA GPUs requires manual container setup, distributed training orchestration, and log parsing. This Skill automates the full evaluation workflow for models like FCN, DeepLabV3, PSPNet, and APCNet on the Cityscapes dataset. ## Core Features & Use Cases - Automated Benchmark Execution: Launches a Docker container, runs batch training via batch_segmentation.sh, and supports configurable GPU counts (1 or 8 cards) with fixed fp16/fp32 precision runs. - Structured Metric Collection: Parses training logs for AVG_ITER_TIME, DATA, and OP timings and writes aggregated results to eval_result.json. - Use Case: Ask the agent to test DeepLabV3 segmentation training performance on 8 NVIDIA GPUs; it starts the container, runs distributed training for both precisions, and returns per-configuration average iteration times. ## Quick Start Ask the agent to test FCN segmentation training performance on NVIDIA GPUs with the Cityscapes dataset and report the average iteration time.

Frequently Asked Questions about nvidia-cv-segmentation

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark segmentation model training on NVIDIA GPUs?▼

Run the batch_segmentation.sh script inside the provided nvidia-cv Docker container with CV_SEG_MODELS and CV_SEG_NGPU set. It launches distributed training via torch.distributed.launch and writes per-configuration results to /workspace/logs/eval_result.json.

Which segmentation models does this benchmark support?▼

The benchmark supports FCN, DeepLabV3, PSPNet, and APCNet, all using ResNet-50 backbones on the Cityscapes dataset at 512x1024 resolution. The default model is FCN, selectable via the CV_SEG_MODELS environment variable.

Can I run the segmentation benchmark with only fp16 precision?▼

No. Precision is fixed to run both fp16 and fp32 for every model and cannot be changed through user input or environment variables. Each precision run produces its own log directory and entry in eval_result.json.

Why does the benchmark fail to find the mmsegmentation source directory?▼

The script requires a directory containing tools/train.py and configs/. It searches common paths automatically, but if your source lives elsewhere, set CV_SEG_MMSEG_DIR explicitly to the onedl-mmsegmentation directory.

What performance metrics does the segmentation benchmark report?▼

It reports avg_iter_time as the core metric, plus data_time for data loading and op_time for computation, parsed from AVG_ITER_TIME log lines. All combinations are aggregated into /workspace/logs/eval_result.json.

Related Skills

Explore More Software Engineering Skills→
kaivyykaivyy
20

download-manager

Download files from URLs into workspace folders with optional SHA-256 verification.

Community
Intermediate
KerberosClawKerberosClaw
77

llm-benchmark

Benchmark local Ollama models across contexts and report performance.

Community
Intermediate
OwnLabAIOwnLabAI
21

torch-geometric

Train graph neural networks using PyG templates, datasets, and benchmarks.

Official
Intermediate
dongg622dongg622

torch_npu

Integrate Huawei Ascend NPU hardware with PyTorch for training and inference.

Community
Advanced