Search HF Hub for aerial segmentation dataset and RF-DETR model
Session from hf-find-traces on Hugging Face by merve · Apache-2.0 · condensed by Coders Talk
Click a segment to jump to that moment
2msession
0interventions
GOAL
Find a commercially permissive real-time object detection model and an aerial satellite image segmentation dataset, then set up a training job for RF-DETR-seg-medium on the iSAID dataset.
- +0Prompt · +0find an aerial satellite image segmentation dataset
- +0mAgent did · +0mRan hf find datasets search and returned a table of aerial segmentation datasets including isaaccorley/isaid and IGNF/FLAIR-1-2
- +0mPrompt · +0mfind a commercially permissive object detection model that's also real-time
- +0mAgent did · +0mSearched hf find models for real-time detectors, hit an invalid 'hf models card' command error, switched to 'hf models info', and compared RT-DETR (Apache 2.0, 43M params) vs YOLOS-tiny (Apache 2.0, 6.5M params) for license and size
- +0mFail · +0mTried 'hf models card' command which doesn't exist
- +1mPrompt · +1mkickoff a job to train RF-DETR-seg-medium on isaid dataset
- +1mAgent did · +1mLoaded the huggingface-vision-trainer skill, checked isaid dataset info, searched for RF-DETR segmentation variants, hit an invalid --files flag error, and located stevenbucaille/rf-detr-seg-medium model info
- +1mFail · +1mTried 'hf datasets info --files' flag which doesn't exist
- +1mAgent did · +1mRan multiple repeated searches for iSAID/aerial segmentation datasets and inspected dataset card metadata and file listings, without landing on a clear confirmed match beyond the original isaaccorley/isaid result
- +2mOutcome · +2mSession ended mid-investigation: model (stevenbucaille/rf-detr-seg-medium) and dataset (isaaccorley/isaid) identified but no training job actually launched yet
VERDICT
Next time I'd specify the exact dataset and model names upfront since the search tool kept returning near-misses and wasted several calls. The agent recovered fine on its own from the two CLI flag errors (hf models card, --files), those weren't real blockers, just syntax mismatches it fixed itself. The session ran out before actually kicking off the training job, so I'd need to follow up to confirm the launch.