Upload folder using huggingface_hub
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +0 -0
- OVERFIT_TROUBLESHOOTING.md +115 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/custom_checkpoint_0.pkl +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/model.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/optimizer.bin +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/random_states_0.pkl +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/scheduler.bin +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/trainer_state.json +1 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/weights.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/custom_checkpoint_0.pkl +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/model.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/optimizer.bin +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/random_states_0.pkl +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/scheduler.bin +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/trainer_state.json +1 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/weights.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/custom_checkpoint_0.pkl +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/model.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/optimizer.bin +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/random_states_0.pkl +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/scheduler.bin +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/trainer_state.json +1 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/weights.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/converted_weights.safetensors +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/debug.log +10 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/ref_0/id0_img0.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/ref_0/id0_img1.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/ref_0/id0_img2.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/debug.log +10 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/ref_0/id0_img0.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/ref_0/id0_img1.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/ref_0/id0_img2.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/debug.log +5 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/ref_0/id0_img0.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/ref_0/id0_img1.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/ref_0/id0_img2.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/debug.log +5 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/ref_0/id0_img0.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/ref_0/id0_img1.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/ref_0/id0_img2.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/debug.log +5 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/ref_0/id0_img0.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/ref_0/id0_img1.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/ref_0/id0_img2.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/debug.log +5 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/ref_0/id0_img0.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/ref_0/id0_img1.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/ref_0/id0_img2.png +3 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_103/debug.log +5 -0
- multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_103/ref_0/id0_img0.png +3 -0
.gitattributes
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
OVERFIT_TROUBLESHOOTING.md
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Overfit 排查清单(从数据 → 训练 → 推理)
|
| 2 |
+
|
| 3 |
+
下面按“最可能导致结果糊”的问题优先级排序。每一条都附带定位方式和建议修复思路。
|
| 4 |
+
|
| 5 |
+
## 1. 数据与标注是否真的被模型用上
|
| 6 |
+
|
| 7 |
+
### 1.1 文本 prompt 在训练与推理不一致(高优先级)
|
| 8 |
+
- 现状:
|
| 9 |
+
- 训练时 `pre_shot_caption` 被固定为 `["xxx", "xxx", "xxx"]`
|
| 10 |
+
- 推理时 `overfit_infer_debug.py` 会读取 `dataset.json` 的 `text` 作为 prompt
|
| 11 |
+
- 影响:模型训练时从未见过真实文本,推理时给真实文本会导致条件不匹配,输出容易发散或模糊。
|
| 12 |
+
- 定位:
|
| 13 |
+
- `multi-shot/multi_view/datasets/videodataset.py` 的 `__getitem__` 硬编码 `pre_shot_caption`
|
| 14 |
+
- 建议:
|
| 15 |
+
- 训练时用真实 `text`(或按 shot 拆分后的文本)替换 `pre_shot_caption`
|
| 16 |
+
- 或推理时也用 `"xxx"` 保持一致(不推荐)
|
| 17 |
+
|
| 18 |
+
### 1.2 overfit 推理只跑了 test split(高优先级)
|
| 19 |
+
- 现状:
|
| 20 |
+
- `overfit_infer_debug.py` 里 `training=False`,Dataset 只使用 `data_test`
|
| 21 |
+
- 你的日志里 `Dataset size: 1`,说明只在 test 上推理
|
| 22 |
+
- 影响:过拟合训练是在 `data_train` 上,推理却在 test 上,必然看不到“过拟合效果”
|
| 23 |
+
- 定位:
|
| 24 |
+
- `MulltiShot_MultiView_Dataset(..., training=False)`
|
| 25 |
+
- 建议:
|
| 26 |
+
- overfit 验证时改为 `training=True` 或直接读取全部数据(不做 train/test split)
|
| 27 |
+
|
| 28 |
+
### 1.3 每次取随机 5 秒片段(高优先级)
|
| 29 |
+
- 现状:
|
| 30 |
+
- `load_video_crop_ref_image` 每次随机切 5 秒,训练/推理都不确定
|
| 31 |
+
- 影响:同一个 index 每次取到不同片段,过拟合目标不稳定 → 输出糊
|
| 32 |
+
- 建议:
|
| 33 |
+
- overfit 模式下固定切片(比如固定从第 0 帧开始)
|
| 34 |
+
- 或设置固定随机种子,并在推理里复用同一个切片
|
| 35 |
+
|
| 36 |
+
## 2. 数据预处理是否正确(process_data)
|
| 37 |
+
|
| 38 |
+
### 2.1 检查处理后的视频尺寸/帧率
|
| 39 |
+
```bash
|
| 40 |
+
python - <<'PY'
|
| 41 |
+
import json, imageio
|
| 42 |
+
from pathlib import Path
|
| 43 |
+
|
| 44 |
+
path = Path('/data/rczhang/PencilFolder/multi-shot/processed/dataset.json')
|
| 45 |
+
data = json.load(path.open())
|
| 46 |
+
first = next(iter(data.values()))
|
| 47 |
+
video = first['disk_path']
|
| 48 |
+
reader = imageio.get_reader(video)
|
| 49 |
+
meta = reader.get_meta_data()
|
| 50 |
+
frame = reader.get_data(0)
|
| 51 |
+
reader.close()
|
| 52 |
+
print('video:', video)
|
| 53 |
+
print('fps:', meta.get('fps'))
|
| 54 |
+
print('shape:', frame.shape) # 期望 (480, 832, 3)
|
| 55 |
+
PY
|
| 56 |
+
```
|
| 57 |
+
- 期望:fps=16,shape=(480, 832, 3)
|
| 58 |
+
|
| 59 |
+
### 2.2 检查 ID 图是否存在、数量够不够
|
| 60 |
+
```bash
|
| 61 |
+
python - <<'PY'
|
| 62 |
+
import json, os
|
| 63 |
+
from pathlib import Path
|
| 64 |
+
|
| 65 |
+
path = Path('/data/rczhang/PencilFolder/multi-shot/processed/dataset.json')
|
| 66 |
+
data = json.load(path.open())
|
| 67 |
+
first = next(iter(data.values()))
|
| 68 |
+
id_dir = Path(first['id_dir'])
|
| 69 |
+
print('id_dir exists:', id_dir.exists())
|
| 70 |
+
print('id_names:', first.get('id_names'))
|
| 71 |
+
for name in first.get('id_names', []):
|
| 72 |
+
imgs = list((id_dir / name).glob('*'))
|
| 73 |
+
print(name, 'num_imgs:', len(imgs))
|
| 74 |
+
PY
|
| 75 |
+
```
|
| 76 |
+
|
| 77 |
+
## 3. 训练配置是否支持“过拟合检查”
|
| 78 |
+
|
| 79 |
+
### 3.1 训练是否在用 ref_images
|
| 80 |
+
- `train_args.zero_face_ratio`=0.1 会随机把 ref 清空 10% 步。
|
| 81 |
+
- 过拟合排查建议:临时设为 `0.0`,确保 ID 图一直生效。
|
| 82 |
+
|
| 83 |
+
### 3.2 训练是否从旧 checkpoint 继续
|
| 84 |
+
- `resume_from_checkpoint: True`
|
| 85 |
+
- 如果之前的数据或 prompt 变了,旧 ckpt 会污染。
|
| 86 |
+
- 建议:排查时用新的输出目录或手动关闭 resume。
|
| 87 |
+
|
| 88 |
+
## 4. 推理配置是否合理
|
| 89 |
+
|
| 90 |
+
### 4.1 纯 prompt 推理 vs v2v
|
| 91 |
+
- 你现在默认是纯 prompt(`input_video=None`),这是合理的。
|
| 92 |
+
- 如果要确认模型是否“记住”训练样本,可临时开 v2v:
|
| 93 |
+
- `--use_input_video`
|
| 94 |
+
- 这只用于 sanity check,不是最终推理形态。
|
| 95 |
+
|
| 96 |
+
### 4.2 采样步数过少
|
| 97 |
+
- `num_inference_steps=8` 非常少,容易糊。
|
| 98 |
+
- 排查建议:先用 20~50 步看看清晰度提升。
|
| 99 |
+
|
| 100 |
+
## 5. 模型与权重是否正确
|
| 101 |
+
|
| 102 |
+
### 5.1 确认加载的是你训练出的最新 checkpoint
|
| 103 |
+
- `overfit_infer_debug.py` 默认取最新 `checkpoint-step-*`
|
| 104 |
+
- 建议手动指定 `--checkpoint_path` 再确认一次。
|
| 105 |
+
|
| 106 |
+
### 5.2 只训练了 DiT
|
| 107 |
+
- `trainable_models: dit`,text encoder 和 VAE 都冻结
|
| 108 |
+
- 对于过拟合一般没问题,但如果 prompt 变化很大,text encoder 冻结会让文本条件效果变弱。
|
| 109 |
+
|
| 110 |
+
## 快速结论(最可能导致“全是糊图”的 3 个点)
|
| 111 |
+
1) **训练/推理 prompt 不一致(训练固定 "xxx",推理用真实 text)**
|
| 112 |
+
2) **overfit 推理用的是 test split,不是训练集**
|
| 113 |
+
3) **每次都随机 5 秒片段,过拟合目标不稳定**
|
| 114 |
+
|
| 115 |
+
如果你愿意,我可以按这个顺序逐一修掉并验证。
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/custom_checkpoint_0.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:33034d45fdab3d8a29157ff26696c9715757dd8d98a0303bfc0a6cbf096cc3a6
|
| 3 |
+
size 1333
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7d80e0d4a0e82af3c59d22a1ab4393c381e2c818eeac07d2dfaab5de636181fb
|
| 3 |
+
size 22770922152
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/optimizer.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:77932c7804d268d4a5e86fb931239710f7cda353f376c4e5db55f1bc92ff9730
|
| 3 |
+
size 19999867838
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/random_states_0.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ad6e005396b1bd1f9ffbd8b85ff8201c2403424f3b3fe9e1e3d679a689a0ae49
|
| 3 |
+
size 14757
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/scheduler.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7ebd442f8f9275fbbec110e74dd49ce951cd2f879e41aa4af850720f5f1e1ce1
|
| 3 |
+
size 1465
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/trainer_state.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"global_step": 450}
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-450-epoch-113/weights.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1a342072c8f551c9714b63b9fef4f91c8bf181b7c91378612c1b109c25329ff4
|
| 3 |
+
size 9999659704
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/custom_checkpoint_0.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:16091977b6e9280ab8ccb5c76eb208ad1c0c6a73ea0510dfba8d996c8afdee3f
|
| 3 |
+
size 1333
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:390d4dde8feb0a40d2ded2fb1f70da9a8ea266d955664b269be1e55912afe999
|
| 3 |
+
size 22770922152
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/optimizer.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2b5d9be57ee5306984a8da64b49611be3ed94857cda09e2888021b1fcd02f5dc
|
| 3 |
+
size 19999867838
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/random_states_0.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7e0b2e9d8d1744a3a9a33c5c5a3cf391884fbbe0ed354d373a3c448e8193fe16
|
| 3 |
+
size 14757
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/scheduler.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:93f6dba1b250c9f1bb90971ef51251e8859c3caaa7acf899c528c3bcc225bf25
|
| 3 |
+
size 1465
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/trainer_state.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"global_step": 600}
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-600-epoch-150/weights.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5f41f0639f62fbd730cdd57da91a91b488120ad3ec9e304f2ffb400c9408ed68
|
| 3 |
+
size 9999659704
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/custom_checkpoint_0.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b629bb6ce063826d133bd4e545f8d7d15b8342d89b0e15001901487d4b5ed7e2
|
| 3 |
+
size 1333
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bf6b18f0a7b1daccb31c4a156a88421b49300c9fe8632b4fc0d2b47089bdfc9b
|
| 3 |
+
size 22770922152
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/optimizer.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b1103a0cf882fb16b74572f2575d91f14658f96af68c79f41f35b998378b7c45
|
| 3 |
+
size 19999867838
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/random_states_0.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e231dad9214f4ef95bb8e0955398dcc89a289757330e31697e473c8b2bb784bf
|
| 3 |
+
size 14757
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/scheduler.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7f0df8db4523b5a804e078196e82594f5b6d138fcb8e21f846e7703ddd2831fc
|
| 3 |
+
size 1465
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/trainer_state.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"global_step": 750}
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/checkpoint-step-750-epoch-188/weights.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f7c24186eb803094ae3b5c5952b0f29685953d3f99eaef1f57b5ec4b5a1b17e1
|
| 3 |
+
size 9999659704
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/converted_weights.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:87d5d43bce868daf997c1069295acd8d9689ff2e09a3482eab2c070e7ec5a968
|
| 3 |
+
size 9999659672
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/debug.log
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=0 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
| 6 |
+
================================================================================
|
| 7 |
+
[debug_infer] epoch=0 index=0
|
| 8 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 9 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 10 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/ref_0/id0_img1.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_0/ref_0/id0_img2.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/debug.log
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=1 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
| 6 |
+
================================================================================
|
| 7 |
+
[debug_infer] epoch=1 index=0
|
| 8 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 9 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 10 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/ref_0/id0_img1.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_1/ref_0/id0_img2.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/debug.log
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=10 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/ref_0/id0_img1.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_10/ref_0/id0_img2.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/debug.log
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=100 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/ref_0/id0_img1.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_100/ref_0/id0_img2.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/debug.log
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=101 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/ref_0/id0_img1.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_101/ref_0/id0_img2.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/debug.log
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=102 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/ref_0/id0_img1.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_102/ref_0/id0_img2.png
ADDED
|
Git LFS Details
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_103/debug.log
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
================================================================================
|
| 2 |
+
[debug_infer] epoch=103 index=0
|
| 3 |
+
video_path=/data/rczhang/PencilFolder/multi-shot/processed/Data1/2.mp4
|
| 4 |
+
num_frames=81 ref_num=3 ID_num=1
|
| 5 |
+
prompt=['画面右侧的黄种人中年男性,穿着深色西装外套和白色衬衫,面朝镜头坐在沙发上,身体前倾,目光看向画面左侧的人,随后微微低头,又抬头看向画面左侧的人。画面左侧的人背对镜头,只露出部分头部和肩膀,穿着蓝色衣服,背景中可以看到木质墙面装饰,墙上挂有带有几何图案的装饰画,增添了空间的艺术感。透过窗户,自然光线洒入室内,营造出明亮而温馨的氛围。视频以晃动的运镜手法,在特写上半身的镜头下展示。', '画面中间的黄种人中年男性面向画面左侧,目光专注地注视着前方。他的头部微微晃动。他穿着深色西装和白色衬衫,头发整齐地向后梳。背景是一面浅色墙壁和一扇带有百叶窗的窗户。整个场景光线柔和,营造出一种安静而专业的氛围。该视频以左摇的运镜手法,在特写胸部以上的画面下拍摄']
|
multi_view/ckpts/Wan2.2_5B-Multi_view-normal_rope_480_832-3ref_local/debug_infer/epoch_103/ref_0/id0_img0.png
ADDED
|
Git LFS Details
|