You can deploy this model and start earning money today!

SLynMi/
Building_Footprint_Extraction___claude_footprint_model

Name: Building_Footprint_Extraction___claude_footprint_model
Availability: InStock

Views

No views yet

Building Footprint Extraction — FootprintNet checkpoints

Checkpoints for the BONAI building-footprint extraction project. Each folder holds the best checkpoint for one model run. The model family lineage is V0–V3 (ResNet-50 base), V4-Attention (RegNetY-6.4GF), V5-Attention (18-channel pruned), and V6 (SegFormer mit_b2) which is the current family.

Current best model: v6_8ch

File	Description
`v6_8ch/best_v6_8ch.pt`	Best checkpoint, 8 input channels, mIoU 0.5249 (TTA 0.5317)
`v6_8ch/best_v6_8ch_stripped.pt`	Deployable stripped weights (inference only)
`v6_8ch/eval_summary.txt`	Held-out eval metrics
`v6_8ch/tta_summary.txt`	HFlip+VFlip TTA comparison

Best config: in_channels=8, SegFormer mit_b2, batch_size=24, accum_steps=2, lr=3e-5, warmup_epochs=8, early_stop_patience=30, grad_checkpoint=true, amp=true.

Model summary

Folder	Input channels	Epoch	mIoU	Status
v6_8ch	8	129	0.5249	Current best
v6_19ch	19	83	0.4977	Deprecated
v6_14ch	14	63	0.5065	Deprecated (Phase 69)
v6_11ch	11	73	0.4958	Deprecated (Phase 70)
v5_attn_1	18	24	0.3867	Deprecated
v5_attn_1_continued	18	35	0.4109	Deprecated
v4_attn_retry_p41	19	23	0.4046	Deprecated
v4_attn_synthetic_roof_p44	19	34	0.4025	Deprecated
channel_attention_phase41_earlystop	—	2	0.3864	Probe (early stop)
channel_attention_phase44_earlystop	—	3	0.3713	Probe (early stop)

SLynMi/Building_Footprint_Extraction___claude_footprint_model

Building Footprint Extraction — FootprintNet checkpoints

Current best model: v6_8ch

Model summary

SLynMi/
Building_Footprint_Extraction___claude_footprint_model