segformer-b0-FT-uff-roi-segmentation-all5-vc1-pc0-cw1-e50-260717-2234

This model is a fine-tuned version of nvidia/mit-b0 on the LeninGF/uff-thermography-multiview-segmentation dataset. It achieves the following results on the evaluation set:

• Loss: 0.0878
• Mean Iou: 0.9213
• Mean Accuracy: 0.9619
• Overall Accuracy: 0.9605
• Accuracy Background: 0.9553
• Accuracy Roi: 0.9685
• Iou Background: 0.9359
• Iou Roi: 0.9067
• Roi Precision: 0.9343
• Roi Recall: 0.9685
• Roi F1: 0.9511
• Roi Dice: 0.9511
• Roi Specificity: 0.9553

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 50

Training results

Framework versions

• Transformers 5.12.1
• Pytorch 2.6.0+cu118
• Datasets 5.0.0
• Tokenizers 0.22.2

Experimental Setup and Reproducibility Configuration

This section is auto-synced from the notebook run configuration.

Run Summary

• saved_at: 2026-07-17T22:34:04.786119-05:00
• run_tag: all5-vc1-pc0-cw1-e50-260717-2234
• dataset: LeninGF/uff-thermography-multiview-segmentation
• selected_view: all_canonical_views
• use_view_conditioning: True. When True, the model receives a view ID embedding to condition predictions on the view.
• use_pseudo_color: False (map=inferno). When True, the model receives a pseudo-colored image instead of the original grayscale image.
• epochs: 50
• learning_rate: 1e-05
• ce_weight/dice_weight: 0.5/0.5
• use_train_augmentation: True. When True, the training dataset is augmented with random transformations to improve generalization.
• aug_prob_hflip: 0.0. If 0.0, horizontal flips are disabled; if 1.0, all training images are flipped; if in (0.0, 1.0), flips occur with the given probability.
• use_class_weights: True. When True, the loss function uses class weights to handle class imbalance at pixel level.

Full Experiment Artifact

• JSON artifact: experiment_conditions.json

JSON Snapshot

{
  "saved_at": "2026-07-17T22:34:04.786119-05:00",
  "dataset": {
    "hf_dataset_identifier": "LeninGF/uff-thermography-multiview-segmentation",
    "shuffle_seed": 42,
    "split_seed": 42
  },
  "view_selection": {
    "selected_view": "all_canonical_views",
    "use_view_conditioning": true
  },
  "augmentation_preprocessing": {
    "use_train_augmentation": true,
    "aug_prob_hflip": 0.0,
    "aug_prob_vflip": 0.0,
    "aug_max_rotation_deg": 5.0,
    "aug_max_translate": 0.03,
    "aug_scale_min": 0.95,
    "aug_scale_max": 1.05,
    "aug_brightness": 0.1,
    "aug_contrast": 0.1,
    "use_pseudo_color": false,
    "pseudo_color_map": "inferno"
  },
  "model": {
    "pretrained_model_name": "nvidia/mit-b0",
    "view_embed_dim": 32
  },
  "training": {
    "epochs": 50,
    "lr": 1e-05,
    "use_dynamic_batch_size": true,
    "batch_size_override": null,
    "use_fp16": false,
    "save_total_limit": 2,
    "eval_strategy": "epoch",
    "save_strategy": "epoch",
    "logging_steps": 10,
    "eval_accumulation_steps": 5,
    "load_best_model_at_end": true,
    "greater_is_better": true,
    "max_grad_norm": 1.0
  },
  "loss_metrics": {
    "IGNORE_INDEX": 255,
    "ce_weight": 0.5,
    "dice_weight": 0.5,
    "use_class_weights": true
  },
  "hub_output": {
    "hf_username": "LeninGF",
    "push_to_hub": true,
    "hub_private_repo": true,
    "hub_strategy": "end",
    "load_from_local_best": true,
    "hub_model_id": "segformer-b0-FT-uff-roi-segmentation-all5-vc1-pc0-cw1-e50-260717-2234",
    "output_dir": "segformer-b0-FT-uff-roi-out-all5-vc1-pc0-cw1-e50-260717-2234",
    "best_model_dir": "segformer-b0-FT-uff-roi-out-all5-vc1-pc0-cw1-e50-260717-2234/best-model",
    "images_dir": "segformer-b0-FT-uff-roi-out-all5-vc1-pc0-cw1-e50-260717-2234/images"
  },
  "naming": {
    "name_base": "segformer-b0-FT-uff-roi-segmentation",
    "view_code": "all5",
    "use_view_conditioning(vc_code)": "vc1",
    "use_pseudo_color(pc_code)": "pc0",
    "class_weights_code": "cw1",
    "exp_code": "all5-vc1-pc0-cw1-e50",
    "run_datetime": "260717-2234",
    "run_tag": "all5-vc1-pc0-cw1-e50-260717-2234",
    "max_name_len": 96,
    "max_tag_len": 36
  },
  "benchmark": {
    "benchmark_n_runs": 5,
    "benchmark_samples_per_run": 20,
    "benchmark_show_first_run_panel": true
  },
  "hub_tags": [
    "vision",
    "image-segmentation",
    "medical-imaging",
    "thermography",
    "multiview",
    "breast-cancer",
    "roi-segmentation"
  ]
}