File size: 1,686 Bytes
ac7cafe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
{
  "schema": "wheat-exp.run-config",
  "schema_version": "1.0.0",
  "experiment": "joint-root-area-poc",
  "run_id": "0003-point18-joint-area-poc",
  "seed": 1001,
  "physical_gpu": 5,
  "logical_device": "cuda:0",
  "image_id": "day2/tiles_png_1001/tile_1001_0_0.png",
  "image_width": 3797,
  "image_height": 3943,
  "root_count": 209,
  "polygon_prediction_count": 401,
  "polygon_array_count": 457,
  "root_xml": "point_id18_annotations.xml",
  "raw_image": "tile_1001_0_0_raw.png",
  "polygon_source": "test_tile_1001_0_0.json",
  "derived_polygon_labelme": "tile_1001_0_0_pseudo_area_labelme.json",
  "backbone": "facebook/dinov3-vitb16-pretrain-lvd1689m",
  "freeze_backbone": true,
  "patch_size": 512,
  "stride": 512,
  "patch_count": 64,
  "train_patch_count": 32,
  "validation_patch_count": 32,
  "split": "row-major edge-push coordinates; first 32 train, last 32 validation",
  "epochs": 200,
  "batch_size": 2,
  "physical_microbatch_size": 1,
  "gradient_accumulation_steps": 2,
  "precision": "CUDA autocast float16; GradScaler; losses accumulated as float32 where PyTorch requires",
  "decoder_activation_checkpointing": true,
  "learning_rate": 0.0001,
  "weight_decay": 0.0001,
  "density_alpha": 0.5,
  "classification_positive_weight": 10.0,
  "classification_sigma": 3.0,
  "use_area": true,
  "area_lambda": 0.15,
  "classification_probability_threshold": 0.5,
  "peak_min_distance": 8,
  "match_tolerance_px": 15.0,
  "preprocessing": "RGB; pixels with all channels >=250 filled black; ImageNet normalization",
  "inference_patch_size": 512,
  "inference_stride": 512,
  "unseen_image": "/mnt/hdb/chenyizi/sam3-fewshot/data/tile_1020_test/tile_1020_0_0.png"
}