Dhruv1000's picture
Organize complete final models, all ablations, logs and checkpoints with visual guides
954bb80 verified
Raw History Blame Contribute Delete
2.88 kB
/usr/lib/python3/dist-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.17.3 and <1.25.0 is required for this version of SciPy (detected version 1.26.4
warnings.warn(f"A NumPy version >={np_minversion} and <{np_maxversion}"
[EVAL] fold [4]: 482 patches (7,712 crop-equiv)
[EVAL] window=128x128 batch=1 -> 482 batches TTA=off ensemble=1 ckpt(s)
[EVAL] /home/ubuntu/tsr/research_round/checkpoints/sam/model_best_raw.tar
epoch=10 val_mIoU=0.5000673383500418 which=raw
/home/ubuntu/tsr/evaluate_test.py:116: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
with autocast():
eval 0/482
eval 50/482
eval 100/482
eval 150/482
eval 200/482
eval 250/482
eval 300/482
eval 350/482
eval 400/482
eval 450/482
==========================================================
T-SRDA v2 evaluation — PASTIS fold-4 test set (official protocol)
482 full 128x128 patches · 3,961,114 scored pixels
TTA: OFF (official) checkpoints: 1
==========================================================
mIoU : 0.4750
OA : 0.7959
mF1 : 0.6060
Kappa : 0.7482
Cls Name IoU F1 Support
----------------------------------------------------------------
1 Meadow 0.8020 0.8901 1,466,003
2 Soft winter wheat 0.7025 0.8253 560,680
3 Corn 0.8569 0.9230 688,289
4 Winter barley 0.4945 0.6617 163,392
5 Winter rapeseed 0.8760 0.9339 129,119
6 Spring barley 0.3614 0.5310 53,762
7 Sunflower 0.4251 0.5966 81,933
8 Grapevine 0.5731 0.7286 187,477
9 Beet 0.7824 0.8779 71,047
10 Winter triticale 0.1223 0.2180 58,362
11 Winter durum wheat 0.5315 0.6941 69,960
12 Fruits, vegetables, flowers 0.2698 0.4250 60,715
13 Potatoes 0.2983 0.4596 22,613
14 Leguminous fodder 0.2718 0.4274 127,794
15 Soybeans 0.5734 0.7289 68,316
16 Orchard 0.2937 0.4541 76,859
17 Mixed cereal 0.0927 0.1697 42,769 *rare*
18 Sorghum 0.2224 0.3638 32,024
Corrected foreground mIoU: 0.47500097242802003
Corrected foreground macro F1: 0.6060348191643534
Saved results: /home/ubuntu/tsr/research_round/checkpoints/sam/test_results.json
peak VRAM: 5.00 GB at batch 1
--- reference points (v1, same protocol, no TTA) ---
v1 fixed epoch 99 : test mIoU 0.4124
v1 best-val ep37 : test mIoU 0.4433
A_linear baseline : test mIoU 0.4747 (seed 3407, best-val)