Spaces:
Runtime error
Runtime error
| configuration: | |
| batch_size: 64 | |
| optimizer: torch.optim.AdamW | |
| lr: 0.001 | |
| trainer: experiment_setup.train_loop | |
| scorer: experiment_setup.score | |
| model: models.clipseg.CLIPDensePredT | |
| lr_scheduler: cosine | |
| T_max: 20000 | |
| eta_min: 0.0001 | |
| max_iterations: 20000 | |
| val_interval: null | |
| # dataset | |
| dataset: datasets.phrasecut.PhraseCut # <----------------- | |
| split_mode: pascal_test | |
| split: train | |
| mask: text_and_crop_blur_highlight352 | |
| image_size: 352 | |
| normalize: True | |
| pre_crop_image_size: [sample, 1, 1.5] | |
| aug: 1new | |
| # general | |
| mix: False # <----------------- | |
| prompt: shuffle+ | |
| norm_cond: True | |
| mix_text_min: 0.0 | |
| # model | |
| out: 1 | |
| extract_layers: [3, 7, 9] | |
| reduce_dim: 64 | |
| depth: 3 | |
| fix_shift: False | |
| loss: torch.nn.functional.binary_cross_entropy_with_logits | |
| amp: True | |
| test_configuration_common: | |
| normalize: True | |
| image_size: 352 | |
| batch_size: 32 | |
| # max_iterations: 5 | |
| # max_iterations: 150 | |
| test_configuration: | |
| - | |
| name: pc # old: phrasecut | |
| metric: metrics.FixedIntervalMetrics | |
| test_dataset: phrasecut | |
| split: test | |
| mask: text | |
| label_support: True | |
| sigmoid: True | |
| columns: [i, name, pc_miou_0.3, pc_fgiou_0.3, pc_fgiou_0.5, pc_ap, duration, date] | |
| individual_configurations: | |
| # important ones | |
| - {name: rd64-uni, version: 'ViT-B/16', reduce_dim: 64, with_visual: True, negative_prob: 0.2, mix: True, mix_text_max: 0.5} | |
| # this was accedentally trained using old mask | |
| - {name: rd128-vit16-phrasecut, version: 'ViT-B/16', reduce_dim: 128, mask: text_and_blur3_highlight01} | |
| - {name: rd64-uni-novis, version: 'ViT-B/16', reduce_dim: 64, with_visual: False, negative_prob: 0.2, mix: False} | |
| # this was accedentally trained using old mask | |
| - {name: baseline3-vit16-phrasecut, model: models.clipseg.CLIPDenseBaseline, version: 'ViT-B/16', reduce_dim: 64, reduce2_dim: 64, mask: text_and_blur3_highlight01} | |
| - {name: vit64-uni, version: 'ViT-B/16', model: models.vitseg.VITDensePredT, reduce_dim: 64, with_visual: True, only_visual: True, negative_prob: 0.2, mask: crop_blur_highlight352, lr: 0.0003} | |
| - {name: vit64-uni-novis, version: 'ViT-B/16', model: models.vitseg.VITDensePredT, with_visual: False, reduce_dim: 64, lr: 0.0001} |