feat: retain best checkpoints and support alternate output roots

configs/drf/drf_scoliosis1k_112_sigma15_joint8_bodyonly_plaince_bridge_eval_2000_1gpu.yaml

@@ -0,0 +1,69 @@
+data_cfg:
+  dataset_name: Scoliosis1K
+  dataset_root: /mnt/public/data/Scoliosis1K/Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly
+  dataset_partition: ./datasets/Scoliosis1K/Scoliosis1K_112.json
+  num_workers: 1
+  remove_no_gallery: false
+
+evaluator_cfg:
+  enable_float16: true
+  restore_ckpt_strict: true
+  restore_hint: 2000
+  save_name: DRF_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k
+  eval_func: evaluate_scoliosis
+  sampler:
+    batch_shuffle: false
+    batch_size: 1
+    sample_type: all_ordered
+    type: InferenceSampler
+    frames_all_limit: 720
+  metric: euc
+  transform:
+    - type: BaseSilCuttingTransform
+    - type: NoOperation
+
+loss_cfg:
+  - loss_term_weight: 1.0
+    margin: 0.2
+    type: TripletLoss
+    log_prefix: triplet
+  - loss_term_weight: 1.0
+    scale: 16
+    type: CrossEntropyLoss
+    log_prefix: softmax
+    log_accuracy: true
+
+model_cfg:
+  model: DRF
+  num_pairs: 8
+  num_metrics: 3
+  backbone_cfg:
+    type: ResNet9
+    block: BasicBlock
+    in_channel: 2
+    channels:
+      - 64
+      - 128
+      - 256
+      - 512
+    layers:
+      - 1
+      - 1
+      - 1
+      - 1
+    strides:
+      - 1
+      - 2
+      - 2
+      - 1
+    maxpool: false
+  SeparateFCs:
+    in_channels: 512
+    out_channels: 256
+    parts_num: 16
+  SeparateBNNecks:
+    class_num: 3
+    in_channels: 256
+    parts_num: 16
+  bin_num:
+    - 16

configs/sconet/sconet_scoliosis1k_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_cosine_finetune_1gpu_80k.yaml

@@ -0,0 +1,115 @@
+data_cfg:
+  dataset_name: Scoliosis1K
+  dataset_root: /mnt/public/data/Scoliosis1K/Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly
+  dataset_partition: ./datasets/Scoliosis1K/Scoliosis1K_112.json
+  data_in_use:
+    - true
+    - false
+  num_workers: 1
+  remove_no_gallery: false
+  test_dataset_name: Scoliosis1K
+
+evaluator_cfg:
+  enable_float16: true
+  restore_ckpt_strict: true
+  restore_hint: 80000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_cosine_finetune_1gpu_80k
+  output_root: /mnt/hddl/data/OpenGait-output
+  eval_func: evaluate_scoliosis
+  sampler:
+    batch_shuffle: false
+    batch_size: 1
+    sample_type: all_ordered
+    type: InferenceSampler
+    frames_all_limit: 720
+  metric: euc
+  transform:
+    - type: BaseSilCuttingTransform
+
+loss_cfg:
+  - loss_term_weight: 1.0
+    margin: 0.2
+    type: TripletLoss
+    log_prefix: triplet
+  - loss_term_weight: 1.0
+    scale: 16
+    type: CrossEntropyLoss
+    log_prefix: softmax
+    log_accuracy: true
+
+model_cfg:
+  model: ScoNet
+  backbone_cfg:
+    type: ResNet9
+    block: BasicBlock
+    in_channel: 2
+    channels:
+      - 64
+      - 128
+      - 256
+      - 512
+    layers:
+      - 1
+      - 1
+      - 1
+      - 1
+    strides:
+      - 1
+      - 2
+      - 2
+      - 1
+    maxpool: false
+  SeparateFCs:
+    in_channels: 512
+    out_channels: 256
+    parts_num: 16
+  SeparateBNNecks:
+    class_num: 3
+    in_channels: 256
+    parts_num: 16
+  bin_num:
+    - 16
+
+optimizer_cfg:
+  lr: 0.0005
+  solver: AdamW
+  weight_decay: 0.0005
+
+scheduler_cfg:
+  scheduler: CosineAnnealingLR
+  T_max: 60000
+  eta_min: 0.00001
+
+trainer_cfg:
+  enable_float16: true
+  fix_BN: false
+  with_test: true
+  log_iter: 100
+  restore_ckpt_strict: true
+  optimizer_reset: false
+  scheduler_reset: false
+  restore_hint: /mnt/hddl/data/OpenGait-output/Scoliosis1K/ScoNet/ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_cosine_finetune_1gpu_80k/checkpoints/latest.pt
+  output_root: /mnt/hddl/data/OpenGait-output
+  auto_resume_latest: true
+  resume_every_iter: 500
+  resume_keep: 6
+  best_ckpt_cfg:
+    keep_n: 3
+    metric_names:
+      - scalar/test_f1/
+      - scalar/test_accuracy/
+  eval_iter: 1000
+  save_iter: 500
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_cosine_finetune_1gpu_80k
+  sync_BN: false
+  total_iter: 80000
+  sampler:
+    batch_shuffle: true
+    batch_size:
+      - 8
+      - 8
+    frames_num_fixed: 30
+    sample_type: fixed_unordered
+    type: TripletSampler
+  transform:
+    - type: BaseSilCuttingTransform

configs/sconet/sconet_scoliosis1k_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_20k.yaml

@@ -0,0 +1,111 @@
+data_cfg:
+  dataset_name: Scoliosis1K
+  dataset_root: /mnt/public/data/Scoliosis1K/Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly
+  dataset_partition: ./datasets/Scoliosis1K/Scoliosis1K_112.json
+  data_in_use:
+    - true
+    - false
+  num_workers: 1
+  remove_no_gallery: false
+  test_dataset_name: Scoliosis1K
+
+evaluator_cfg:
+  enable_float16: true
+  restore_ckpt_strict: true
+  restore_hint: 20000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_20k
+  eval_func: evaluate_scoliosis
+  sampler:
+    batch_shuffle: false
+    batch_size: 1
+    sample_type: all_ordered
+    type: InferenceSampler
+    frames_all_limit: 720
+  metric: euc
+  transform:
+    - type: BaseSilCuttingTransform
+
+loss_cfg:
+  - loss_term_weight: 1.0
+    margin: 0.2
+    type: TripletLoss
+    log_prefix: triplet
+  - loss_term_weight: 1.0
+    scale: 16
+    type: CrossEntropyLoss
+    log_prefix: softmax
+    log_accuracy: true
+
+model_cfg:
+  model: ScoNet
+  backbone_cfg:
+    type: ResNet9
+    block: BasicBlock
+    in_channel: 2
+    channels:
+      - 64
+      - 128
+      - 256
+      - 512
+    layers:
+      - 1
+      - 1
+      - 1
+      - 1
+    strides:
+      - 1
+      - 2
+      - 2
+      - 1
+    maxpool: false
+  SeparateFCs:
+    in_channels: 512
+    out_channels: 256
+    parts_num: 16
+  SeparateBNNecks:
+    class_num: 3
+    in_channels: 256
+    parts_num: 16
+  bin_num:
+    - 16
+
+optimizer_cfg:
+  lr: 0.001
+  solver: AdamW
+  weight_decay: 0.0005
+
+scheduler_cfg:
+  gamma: 0.1
+  milestones:
+    - 5000
+    - 7000
+    - 9000
+  scheduler: MultiStepLR
+
+trainer_cfg:
+  enable_float16: true
+  fix_BN: false
+  with_test: true
+  log_iter: 100
+  restore_ckpt_strict: true
+  optimizer_reset: true
+  scheduler_reset: true
+  restore_hint: output/Scoliosis1K/ScoNet/ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k/checkpoints/latest.pt
+  auto_resume_latest: true
+  resume_every_iter: 500
+  resume_keep: 6
+  eval_iter: 1000
+  save_iter: 1000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_20k
+  sync_BN: false
+  total_iter: 20000
+  sampler:
+    batch_shuffle: true
+    batch_size:
+      - 8
+      - 8
+    frames_num_fixed: 30
+    sample_type: fixed_unordered
+    type: TripletSampler
+  transform:
+    - type: BaseSilCuttingTransform

configs/sconet/sconet_scoliosis1k_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_40k.yaml

@@ -0,0 +1,111 @@
+data_cfg:
+  dataset_name: Scoliosis1K
+  dataset_root: /mnt/public/data/Scoliosis1K/Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly
+  dataset_partition: ./datasets/Scoliosis1K/Scoliosis1K_112.json
+  data_in_use:
+    - true
+    - false
+  num_workers: 1
+  remove_no_gallery: false
+  test_dataset_name: Scoliosis1K
+
+evaluator_cfg:
+  enable_float16: true
+  restore_ckpt_strict: true
+  restore_hint: 40000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_40k
+  eval_func: evaluate_scoliosis
+  sampler:
+    batch_shuffle: false
+    batch_size: 1
+    sample_type: all_ordered
+    type: InferenceSampler
+    frames_all_limit: 720
+  metric: euc
+  transform:
+    - type: BaseSilCuttingTransform
+
+loss_cfg:
+  - loss_term_weight: 1.0
+    margin: 0.2
+    type: TripletLoss
+    log_prefix: triplet
+  - loss_term_weight: 1.0
+    scale: 16
+    type: CrossEntropyLoss
+    log_prefix: softmax
+    log_accuracy: true
+
+model_cfg:
+  model: ScoNet
+  backbone_cfg:
+    type: ResNet9
+    block: BasicBlock
+    in_channel: 2
+    channels:
+      - 64
+      - 128
+      - 256
+      - 512
+    layers:
+      - 1
+      - 1
+      - 1
+      - 1
+    strides:
+      - 1
+      - 2
+      - 2
+      - 1
+    maxpool: false
+  SeparateFCs:
+    in_channels: 512
+    out_channels: 256
+    parts_num: 16
+  SeparateBNNecks:
+    class_num: 3
+    in_channels: 256
+    parts_num: 16
+  bin_num:
+    - 16
+
+optimizer_cfg:
+  lr: 0.001
+  solver: AdamW
+  weight_decay: 0.0005
+
+scheduler_cfg:
+  gamma: 0.1
+  milestones:
+    - 10000
+    - 20000
+    - 30000
+  scheduler: MultiStepLR
+
+trainer_cfg:
+  enable_float16: true
+  fix_BN: false
+  with_test: true
+  log_iter: 100
+  restore_ckpt_strict: true
+  optimizer_reset: true
+  scheduler_reset: true
+  restore_hint: output/Scoliosis1K/ScoNet/ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k/checkpoints/latest.pt
+  auto_resume_latest: true
+  resume_every_iter: 500
+  resume_keep: 6
+  eval_iter: 1000
+  save_iter: 1000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_40k
+  sync_BN: false
+  total_iter: 40000
+  sampler:
+    batch_shuffle: true
+    batch_size:
+      - 8
+      - 8
+    frames_num_fixed: 30
+    sample_type: fixed_unordered
+    type: TripletSampler
+  transform:
+    - type: BaseSilCuttingTransform

configs/sconet/sconet_scoliosis1k_skeleton_112_sigma15_joint8_bodyonly_plaince_main_1gpu_20k.yaml

@@ -0,0 +1,110 @@
+data_cfg:
+  dataset_name: Scoliosis1K
+  dataset_root: /mnt/public/data/Scoliosis1K/Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly
+  dataset_partition: ./datasets/Scoliosis1K/Scoliosis1K_112.json
+  data_in_use:
+    - true
+    - false
+  num_workers: 1
+  remove_no_gallery: false
+  test_dataset_name: Scoliosis1K
+
+evaluator_cfg:
+  enable_float16: true
+  restore_ckpt_strict: true
+  restore_hint: 20000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_main_1gpu_20k
+  eval_func: evaluate_scoliosis
+  sampler:
+    batch_shuffle: false
+    batch_size: 1
+    sample_type: all_ordered
+    type: InferenceSampler
+    frames_all_limit: 720
+  metric: euc
+  transform:
+    - type: BaseSilCuttingTransform
+
+loss_cfg:
+  - loss_term_weight: 1.0
+    margin: 0.2
+    type: TripletLoss
+    log_prefix: triplet
+  - loss_term_weight: 1.0
+    scale: 16
+    type: CrossEntropyLoss
+    log_prefix: softmax
+    log_accuracy: true
+
+model_cfg:
+  model: ScoNet
+  backbone_cfg:
+    type: ResNet9
+    block: BasicBlock
+    in_channel: 2
+    channels:
+      - 64
+      - 128
+      - 256
+      - 512
+    layers:
+      - 1
+      - 1
+      - 1
+      - 1
+    strides:
+      - 1
+      - 2
+      - 2
+      - 1
+    maxpool: false
+  SeparateFCs:
+    in_channels: 512
+    out_channels: 256
+    parts_num: 16
+  SeparateBNNecks:
+    class_num: 3
+    in_channels: 256
+    parts_num: 16
+  bin_num:
+    - 16
+
+optimizer_cfg:
+  lr: 0.1
+  momentum: 0.9
+  solver: SGD
+  weight_decay: 0.0005
+
+scheduler_cfg:
+  gamma: 0.1
+  milestones:
+    - 10000
+    - 14000
+    - 18000
+  scheduler: MultiStepLR
+
+trainer_cfg:
+  enable_float16: true
+  fix_BN: false
+  with_test: true
+  log_iter: 100
+  restore_ckpt_strict: true
+  restore_hint: 0
+  auto_resume_latest: true
+  resume_every_iter: 500
+  resume_keep: 6
+  eval_iter: 1000
+  save_iter: 1000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_main_1gpu_20k
+  sync_BN: false
+  total_iter: 20000
+  sampler:
+    batch_shuffle: true
+    batch_size:
+      - 8
+      - 8
+    frames_num_fixed: 30
+    sample_type: fixed_unordered
+    type: TripletSampler
+  transform:
+    - type: BaseSilCuttingTransform

configs/sconet/sconet_scoliosis1k_skeleton_112_sigma15_joint8_bodyonly_plaince_resume_1gpu_20k.yaml

@@ -0,0 +1,112 @@
+data_cfg:
+  dataset_name: Scoliosis1K
+  dataset_root: /mnt/public/data/Scoliosis1K/Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly
+  dataset_partition: ./datasets/Scoliosis1K/Scoliosis1K_112.json
+  data_in_use:
+    - true
+    - false
+  num_workers: 1
+  remove_no_gallery: false
+  test_dataset_name: Scoliosis1K
+
+evaluator_cfg:
+  enable_float16: true
+  restore_ckpt_strict: true
+  restore_hint: 20000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_resume_1gpu_20k
+  eval_func: evaluate_scoliosis
+  sampler:
+    batch_shuffle: false
+    batch_size: 1
+    sample_type: all_ordered
+    type: InferenceSampler
+    frames_all_limit: 720
+  metric: euc
+  transform:
+    - type: BaseSilCuttingTransform
+
+loss_cfg:
+  - loss_term_weight: 1.0
+    margin: 0.2
+    type: TripletLoss
+    log_prefix: triplet
+  - loss_term_weight: 1.0
+    scale: 16
+    type: CrossEntropyLoss
+    log_prefix: softmax
+    log_accuracy: true
+
+model_cfg:
+  model: ScoNet
+  backbone_cfg:
+    type: ResNet9
+    block: BasicBlock
+    in_channel: 2
+    channels:
+      - 64
+      - 128
+      - 256
+      - 512
+    layers:
+      - 1
+      - 1
+      - 1
+      - 1
+    strides:
+      - 1
+      - 2
+      - 2
+      - 1
+    maxpool: false
+  SeparateFCs:
+    in_channels: 512
+    out_channels: 256
+    parts_num: 16
+  SeparateBNNecks:
+    class_num: 3
+    in_channels: 256
+    parts_num: 16
+  bin_num:
+    - 16
+
+optimizer_cfg:
+  lr: 0.1
+  momentum: 0.9
+  solver: SGD
+  weight_decay: 0.0005
+
+scheduler_cfg:
+  gamma: 0.1
+  milestones:
+    - 10000
+    - 14000
+    - 18000
+  scheduler: MultiStepLR
+
+trainer_cfg:
+  enable_float16: true
+  fix_BN: false
+  with_test: true
+  log_iter: 100
+  restore_ckpt_strict: true
+  optimizer_reset: false
+  scheduler_reset: false
+  restore_hint: output/Scoliosis1K/ScoNet/ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k/checkpoints/latest.pt
+  auto_resume_latest: true
+  resume_every_iter: 500
+  resume_keep: 6
+  eval_iter: 1000
+  save_iter: 1000
+  save_name: ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_resume_1gpu_20k
+  sync_BN: false
+  total_iter: 20000
+  sampler:
+    batch_shuffle: true
+    batch_size:
+      - 8
+      - 8
+    frames_num_fixed: 30
+    sample_type: fixed_unordered
+    type: TripletSampler
+  transform:
+    - type: BaseSilCuttingTransform

docs/5.advanced_usages.md

@@ -86,3 +86,27 @@
 >>     summary_writer.add_video('outs', outs.mean(2).unsqueeze(2), self.iteration)
 >> ```
 > Note that this example requires the [`moviepy`](https://github.com/Zulko/moviepy) package, and hence you should run `pip install moviepy` first.
+
+### Keep Best Checkpoints
+> If you want to retain the strongest evaluation checkpoints instead of relying only on the latest or final save, you can enable best-checkpoint tracking in `trainer_cfg`.
+>
+> Example:
+>> ```yaml
+>> trainer_cfg:
+>>   with_test: true
+>>   eval_iter: 1000
+>>   save_iter: 1000
+>>   best_ckpt_cfg:
+>>     keep_n: 3
+>>     metric_names:
+>>       - scalar/test_f1/
+>>       - scalar/test_accuracy/
+>> ```
+>
+> Behavior:
+> * The normal numbered checkpoints are still written by `save_iter`.
+> * After each eval, the trainer checks the configured scalar metrics and keeps the top `N` checkpoints separately for each metric.
+> * Best checkpoints are saved under `output/.../checkpoints/best/<metric>/`.
+> * Each best-metric directory contains an `index.json` file with the retained iterations, scores, and paths.
+>
+> This is useful for long or unstable runs where the best checkpoint may appear well before the final iteration.

docs/scoliosis_reproducibility_audit.md

@@ -164,6 +164,7 @@ Conclusion:
 - on the same split, `body-only + plain CE` improved that further to `83.16 Acc / 68.24 Prec / 80.02 Rec / 68.47 F1` at `7000`
 - a later explicit rerun of the `body-only + plain CE` `7000` full-test eval reproduced that same `83.16 / 68.24 / 80.02 / 68.47` result
 - adding back limited head context via `head-lite` did not improve the full-test score; its `7000` checkpoint reached only `78.07 Acc / 65.42 Prec / 80.50 Rec / 62.08 F1`
+- the first practical DRF bridge on the same `1:1:2` body-only recipe peaked early and still underperformed the plain skeleton baseline; its best retained `2000` checkpoint reached only `80.21 Acc / 58.92 Prec / 59.23 Rec / 57.84 F1` on the full test set
 
 ### Not reproducible with current evidence
 
@@ -179,6 +180,10 @@ Conclusion:
 - the `1:1:8` class ratio is not just a nuisance; it appears to be a major driver of the current skeleton/DRF failure mode
 - on the easier `1:1:2` split, weighted CE is not currently the winning recipe; the best local full-test result so far came from plain CE
 - `head-lite` may help the small fixed proxy subset, but that gain did not transfer to the full `TEST_SET`, so `body-only + plain CE` remains the best practical skeleton recipe
+- DRF currently looks worse than the plain skeleton baseline not because the skeleton path is dead, but because the additional prior branch is not yet providing a selective or stable complement. The current local evidence points to three likely causes:
+  - the body-only skeleton baseline already captures most of the useful torso signal on `1:1:2`, so PAV may be largely redundant in this setting
+  - the current PGA/PAV path appears weakly selective in local diagnostics, so the prior is not clearly emphasizing a few clinically relevant parts
+  - DRF peaks very early and then degrades, which suggests the added branch is making optimization less stable without improving the final decision boundary
 
 ## Recommended standard for future work in this repo
 

docs/scoliosis_training_change_log.md

@@ -37,7 +37,11 @@ Use it for:
 | 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_2gpu_10k` | ScoNet-MT-ske bridge | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | Same `1:1:2` body-only bridge as above, but removed weighted CE to test whether class weighting was suppressing precision on the easier split | interrupted | superseded before meaningful progress by the user-requested 1-GPU rerun on the `5070 Ti` |
 | 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k` | ScoNet-MT-ske bridge | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | Same plain-CE `1:1:2` bridge, relaunched on the `5070 Ti` only per user request | complete | best proxy subset at `7000`: `88.28/69.12/74.15/68.80`; full test at `7000`: `83.16/68.24/80.02/68.47`; final proxy at `10000`: `75.00/65.00/63.41/54.55` (Acc/Prec/Rec/F1) |
 | 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_headlite_plaince_bridge_1gpu_10k` | ScoNet-MT-ske bridge | `Scoliosis1K-drf-pkl-112-sigma15-joint8-headlite` + `Scoliosis1K_112.json` | Added `head-lite` structure (nose plus shoulder links, no eyes/ears) on top of the plain-CE `1:1:2` bridge; first `3090` launch OOMed due unrelated occupancy, then relaunched on the UUID-pinned `5070 Ti` | complete | best proxy subset at `7000`: `86.72/70.15/89.00/70.44`; full test at `7000`: `78.07/65.42/80.50/62.08` (Acc/Prec/Rec/F1) |
-| 2026-03-10 | `DRF_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k` | DRF bridge | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | First practical DRF run on the winning `1:1:2` skeleton recipe: `body-only`, plain CE, SGD, `10k` bridge schedule, fixed proxy subset seed `112` | running | pending |
+| 2026-03-10 | `DRF_skeleton_112_sigma15_joint8_bodyonly_plaince_bridge_1gpu_10k` | DRF bridge | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | First practical DRF run on the winning `1:1:2` skeleton recipe: `body-only`, plain CE, SGD, `10k` bridge schedule, fixed proxy subset seed `112` | complete | best proxy subset at `2000`: `88.28/61.79/60.31/60.93`; full test at `2000`: `80.21/58.92/59.23/57.84` (Acc/Prec/Rec/F1) |
+| 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_main_1gpu_20k` | ScoNet-MT-ske mainline | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | Promoted the winning practical skeleton recipe to a longer `20k` run with full `TEST_SET` eval and checkpoint save every `1000`; no proxy subset, same plain CE + SGD setup | interrupted | superseded by the true-resume continuation below |
+| 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_resume_1gpu_20k` | ScoNet-MT-ske mainline | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | True continuation of the earlier plain-CE `1:1:2` `10k` bridge from its `latest.pt`, extended to `20k` with full `TEST_SET` eval and checkpoint save every `1000` | interrupted | superseded by the AdamW finetune branch below |
+| 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_20k` | ScoNet-MT-ske finetune | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | AdamW finetune from the earlier plain-CE `1:1:2` `10k` checkpoint; restores model weights only, resets optimizer/scheduler state, keeps full `TEST_SET` eval and checkpoint save every `1000` | interrupted | superseded by the longer overnight 40k finetune below |
+| 2026-03-10 | `ScoNet_skeleton_112_sigma15_joint8_bodyonly_plaince_adamw_finetune_1gpu_40k` | ScoNet-MT-ske finetune | `Scoliosis1K-drf-pkl-118-sigma15-joint8-bodyonly` + `Scoliosis1K_112.json` | Longer overnight AdamW finetune from the same `10k` plain-CE checkpoint; restores model weights only, resets optimizer/scheduler state, extends to `40000` total iterations with full `TEST_SET` eval every `1000` | running | pending |
 
 ## Current best skeleton baseline
 
@@ -63,3 +67,4 @@ Current best `ScoNet-MT-ske`-style result:
 - Removing weighted CE on the `1:1:2` bridge improved the current best full-test result further: `body-only + plain CE` reached `83.16 Acc / 68.24 Prec / 80.02 Rec / 68.47 F1` at `7000`, so weighted CE does not currently look beneficial on the easier split.
 - A later full-test rerun of the retained `body-only + plain CE` `7000` checkpoint reproduced the same `83.16 / 68.24 / 80.02 / 68.47` result exactly, so that number is now explicitly reconfirmed rather than just carried forward from the original run log.
 - `Head-lite` looked stronger than `body-only` on the fixed 128-sequence proxy subset at `7000`, but it did not transfer to the full test set: `78.07 Acc / 65.42 Prec / 80.50 Rec / 62.08 F1`, which is clearly below the `body-only + plain CE` full-test result.
+- The first practical DRF bridge on the winning `1:1:2` recipe did not beat the plain skeleton baseline. Its best retained checkpoint (`2000`) reached only `80.21 Acc / 58.92 Prec / 59.23 Rec / 57.84 F1` on the full test set, versus `83.16 / 68.24 / 80.02 / 68.47` for `body-only + plain CE` at `7000`. The working local interpretation is that the added PAV/PGA path is currently injecting a weak or noisy prior rather than a useful complementary signal.

docs/systemd-run-training.md

@@ -124,6 +124,33 @@ The launcher configures both:
 
 This makes it easier to recover logs even if the original shell or tool session disappears.
 
+## Moving outputs off the SSD
+
+OpenGait writes checkpoints, TensorBoard summaries, best-checkpoint snapshots, and file logs under a run output root.
+
+By default that root is `output/`, but you can override it per run with `output_root` in the engine config:
+
+```yaml
+trainer_cfg:
+  output_root: /mnt/hddl/data/OpenGait-output
+
+evaluator_cfg:
+  output_root: /mnt/hddl/data/OpenGait-output
+```
+
+The final path layout stays the same under that root:
+
+```text
+<output_root>/<dataset>/<model>/<save_name>/
+```
+
+For long scoliosis runs, using an HDD-backed root is recommended so local SSD space is not consumed by:
+
+- numbered checkpoints
+- rolling resume checkpoints
+- best-N retained checkpoints
+- TensorBoard summary files
+
 ## GPU selection
 
 Prefer GPU UUIDs, not ordinal indices.

opengait/main.py

@@ -4,7 +4,14 @@ import argparse
 import torch
 import torch.nn as nn
 from modeling import models
-from opengait.utils import config_loader, get_ddp_module, init_seeds, params_count, get_msg_mgr
+from opengait.utils import (
+    config_loader,
+    get_ddp_module,
+    get_msg_mgr,
+    init_seeds,
+    params_count,
+    resolve_output_path,
+)
 
 parser = argparse.ArgumentParser(description='Main program for opengait.')
 parser.add_argument('--local_rank', type=int, default=0,
@@ -25,8 +32,7 @@ def initialization(cfgs, training):
     msg_mgr = get_msg_mgr()
     engine_cfg = cfgs['trainer_cfg'] if training else cfgs['evaluator_cfg']
     logger_cfg = cfgs.get('logger_cfg', {})
-    output_path = os.path.join('output/', cfgs['data_cfg']['dataset_name'],
+    output_path = resolve_output_path(cfgs, engine_cfg)
-                               cfgs['model_cfg']['model'], engine_cfg['save_name'])
     if training:
         msg_mgr.init_manager(
             output_path,

opengait/modeling/base_model.py

@@ -10,8 +10,10 @@ BaseModel.run_train(model)
 BaseModel.run_test(model)
 """
 import json
+import math
 import os
 import random
+import re
 from typing import Any
 
 import numpy as np
@@ -33,7 +35,7 @@ from data.transform import get_transform
 from data.collate_fn import CollateFn
 from data.dataset import DataSet
 import data.sampler as Samplers
-from opengait.utils import Odict, mkdir, ddp_all_gather
+from opengait.utils import Odict, mkdir, ddp_all_gather, resolve_output_path
 from opengait.utils import get_valid_args, is_list, is_dict, np2var, ts2np, list2var, get_attr_from
 from evaluation import evaluator as eval_functions
 from opengait.utils import NoOp
@@ -144,8 +146,7 @@ class BaseModel(MetaModel, nn.Module):
 
         if training and self.engine_cfg['enable_float16']:
             self.Scaler = GradScaler()
-        self.save_path = osp.join('output/', cfgs['data_cfg']['dataset_name'],
+        self.save_path = resolve_output_path(cfgs, self.engine_cfg)
-                                  cfgs['model_cfg']['model'], self.engine_cfg['save_name'])
 
         self.build_network(cfgs['model_cfg'])
         self.init_parameters()
@@ -317,6 +318,134 @@ class BaseModel(MetaModel, nn.Module):
                 return candidate
         return None
 
+    def _best_ckpt_cfg(self) -> dict[str, Any] | None:
+        best_ckpt_cfg = self.engine_cfg.get('best_ckpt_cfg')
+        if not isinstance(best_ckpt_cfg, dict):
+            return None
+        keep_n = int(best_ckpt_cfg.get('keep_n', 0))
+        metric_names = best_ckpt_cfg.get('metric_names', [])
+        if keep_n <= 0 or not isinstance(metric_names, list) or not metric_names:
+            return None
+        return best_ckpt_cfg
+
+    def _best_ckpt_root(self) -> str:
+        return osp.join(self._checkpoint_dir(), "best")
+
+    def _best_metric_dir(self, metric_name: str) -> str:
+        metric_slug = re.sub(r"[^A-Za-z0-9_.-]+", "_", metric_name).strip("._")
+        return osp.join(self._best_ckpt_root(), metric_slug)
+
+    def _best_metric_index_path(self, metric_name: str) -> str:
+        return osp.join(self._best_metric_dir(metric_name), "index.json")
+
+    def _load_best_metric_index(self, metric_name: str) -> list[dict[str, Any]]:
+        index_path = self._best_metric_index_path(metric_name)
+        if not osp.isfile(index_path):
+            return []
+        with open(index_path, "r", encoding="utf-8") as handle:
+            raw_entries = json.load(handle)
+        if not isinstance(raw_entries, list):
+            return []
+        entries: list[dict[str, Any]] = []
+        for entry in raw_entries:
+            if not isinstance(entry, dict):
+                continue
+            path = entry.get("path")
+            if isinstance(path, str) and osp.isfile(path):
+                entries.append(entry)
+        return entries
+
+    def _write_best_metric_index(
+        self,
+        metric_name: str,
+        entries: list[dict[str, Any]],
+    ) -> None:
+        index_path = self._best_metric_index_path(metric_name)
+        mkdir(osp.dirname(index_path))
+        tmp_path = index_path + ".tmp"
+        with open(tmp_path, "w", encoding="utf-8") as handle:
+            json.dump(entries, handle, indent=2, sort_keys=True)
+        os.replace(tmp_path, index_path)
+
+    def _summary_scalar(self, value: Any) -> float | None:
+        if isinstance(value, torch.Tensor):
+            return float(value.detach().float().mean().item())
+        if isinstance(value, np.ndarray):
+            return float(np.mean(value))
+        if isinstance(value, (float, int, np.floating, np.integer)):
+            return float(value)
+        return None
+
+    def _save_best_ckpts(
+        self,
+        iteration: int,
+        result_dict: dict[str, Any],
+    ) -> None:
+        if torch.distributed.get_rank() != 0:
+            return
+        best_ckpt_cfg = self._best_ckpt_cfg()
+        if best_ckpt_cfg is None:
+            return
+
+        keep_n = int(best_ckpt_cfg['keep_n'])
+        metric_names = [metric for metric in best_ckpt_cfg['metric_names'] if metric in result_dict]
+        if not metric_names:
+            return
+
+        checkpoint: dict[str, Any] | None = None
+        save_name = self.engine_cfg['save_name']
+
+        for metric_name in metric_names:
+            score = self._summary_scalar(result_dict.get(metric_name))
+            if score is None or not math.isfinite(score):
+                continue
+
+            entries = [
+                entry for entry in self._load_best_metric_index(metric_name)
+                if int(entry.get("iteration", -1)) != iteration
+            ]
+            ranked_entries = sorted(
+                entries + [{"iteration": iteration, "score": score, "path": ""}],
+                key=lambda entry: (float(entry["score"]), int(entry["iteration"])),
+                reverse=True,
+            )
+            kept_entries = ranked_entries[:keep_n]
+            if not any(int(entry["iteration"]) == iteration for entry in kept_entries):
+                continue
+
+            metric_dir = self._best_metric_dir(metric_name)
+            mkdir(metric_dir)
+            metric_slug = osp.basename(metric_dir)
+            best_path = osp.join(
+                metric_dir,
+                f"{save_name}-iter-{iteration:0>5}-score-{score:.4f}-{metric_slug}.pt",
+            )
+
+            if checkpoint is None:
+                checkpoint = self._build_checkpoint(iteration)
+            self._save_checkpoint_file(checkpoint, best_path)
+
+            refreshed_entries = []
+            for entry in kept_entries:
+                if int(entry["iteration"]) == iteration:
+                    refreshed_entries.append(
+                        {
+                            "iteration": iteration,
+                            "score": score,
+                            "path": best_path,
+                        }
+                    )
+                else:
+                    refreshed_entries.append(entry)
+
+            keep_paths = {entry["path"] for entry in refreshed_entries if isinstance(entry.get("path"), str)}
+            for stale_entry in entries:
+                stale_path = stale_entry.get("path")
+                if isinstance(stale_path, str) and stale_path not in keep_paths and osp.isfile(stale_path):
+                    os.remove(stale_path)
+
+            self._write_best_metric_index(metric_name, refreshed_entries)
+
     def save_ckpt(self, iteration):
         if torch.distributed.get_rank() == 0:
             save_name = self.engine_cfg['save_name']
@@ -589,6 +718,7 @@ class BaseModel(MetaModel, nn.Module):
                 if result_dict:
                     model.msg_mgr.write_to_tensorboard(result_dict)
                     model.msg_mgr.write_to_wandb(result_dict)
+                    model._save_best_ckpts(model.iteration, result_dict)
                 model.msg_mgr.reset_time()
             if model.iteration >= model.engine_cfg['total_iter']:
                 break

opengait/utils/__init__.py

@@ -7,4 +7,5 @@ from .common import mkdir, clones
 from .common import MergeCfgsDict
 from .common import get_attr_from
 from .common import NoOp
+from .common import resolve_output_path
 from .msg_manager import get_msg_mgr

opengait/utils/common.py

@@ -2,6 +2,7 @@ import copy
 import os
 import inspect
 import logging
+from pathlib import Path
 import torch
 import numpy as np
 import torch.nn as nn
@@ -203,3 +204,19 @@ def get_ddp_module(module, find_unused_parameters=False, **kwargs):
 def params_count(net):
     n_parameters = sum(p.numel() for p in net.parameters())
     return 'Parameters Count: {:.5f}M'.format(n_parameters / 1e6)
+
+
+def resolve_output_path(cfgs, engine_cfg):
+    output_root = (
+        engine_cfg.get('output_root')
+        or cfgs.get('output_root')
+        or os.environ.get('OPENGAIT_OUTPUT_ROOT')
+        or 'output'
+    )
+    output_root = str(Path(output_root).expanduser())
+    return os.path.join(
+        output_root,
+        cfgs['data_cfg']['dataset_name'],
+        cfgs['model_cfg']['model'],
+        engine_cfg['save_name'],
+    )