init

README.md

@@ -0,0 +1,40 @@
+## Environment Setup
+
+`pip install -r requirements.txt`
+
+## Download checkpoints
+
+1. Download the pretrained checkpoints of [SVD_xt](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt-1-1) from huggingface to `./ckpts`.
+
+2. Download the checkpint of [MOFA-Adapter](https://huggingface.co/MyNiuuu/MOFA-Video-Traj) from huggingface to `./ckpts`.
+
+The final structure of checkpoints should be:
+
+
+```text
+./ckpts/
+|-- controlnet
+|   |-- config.json
+|   `-- diffusion_pytorch_model.safetensors
+|-- stable-video-diffusion-img2vid-xt-1-1
+|   |-- feature_extractor
+|       |-- ...
+|   |-- image_encoder
+|       |-- ...
+|   |-- scheduler
+|       |-- ...
+|   |-- unet
+|       |-- ...
+|   |-- unet_ch9
+|       |-- ...
+|   |-- vae
+|       |-- ...
+|   |-- svd_xt_1_1.safetensors
+|   `-- model_index.json
+```
+
+## Run Gradio Demo
+
+`python run_gradio.py`
+
+Please refer to the instructions on the gradio interface during the inference process.

ckpts/controlnet/config.json

@@ -0,0 +1,45 @@
+{
+  "_class_name": "FlowControlNet",
+  "_diffusers_version": "0.25.1",
+  "_name_or_path": "/apdcephfs_cq10/share_1290939/myniu/svd_controlnet/svdxt11_featureflow_forward_avg_256256_stride4/unimatch_512384/checkpoint-100000/controlnet",
+  "addition_time_embed_dim": 256,
+  "block_out_channels": [
+    320,
+    640,
+    1280,
+    1280
+  ],
+  "conditioning_channels": 3,
+  "conditioning_embedding_out_channels": [
+    16,
+    32,
+    96,
+    256
+  ],
+  "cross_attention_dim": 1024,
+  "down_block_types": [
+    "CrossAttnDownBlockSpatioTemporal",
+    "CrossAttnDownBlockSpatioTemporal",
+    "CrossAttnDownBlockSpatioTemporal",
+    "DownBlockSpatioTemporal"
+  ],
+  "in_channels": 8,
+  "layers_per_block": 2,
+  "num_attention_heads": [
+    5,
+    10,
+    10,
+    20
+  ],
+  "num_frames": 25,
+  "out_channels": 4,
+  "projection_class_embeddings_input_dim": 768,
+  "sample_size": null,
+  "transformer_layers_per_block": 1,
+  "up_block_types": [
+    "UpBlockSpatioTemporal",
+    "CrossAttnUpBlockSpatioTemporal",
+    "CrossAttnUpBlockSpatioTemporal",
+    "CrossAttnUpBlockSpatioTemporal"
+  ]
+}

ckpts/controlnet/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1303192a1e72d071e15e7eb37fd1ea15f6424aaf2cd6b6b1e1bb3b1e9e75d37e
+size 2777345452

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/config.yaml

@@ -0,0 +1,59 @@
+model:
+    arch: CMP
+    total_iter: 140000
+    lr_steps: [80000, 120000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: alexnet_fcn_32x
+        sparse_encoder: shallownet32x
+        flow_decoder: MotionDecoderPlain
+        skip_layer: False
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 12
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 416
+    crop_size: [384, 384]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 0.000025
+    nms_ks: 81
+    max_num_guide: 150
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/yfcc/lists/train.txt
+        - data/youtube9000/lists/train.txt
+    val_source:
+        - data/yfcc/lists/val.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 10000
+    save_freq: 10000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: False

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/resume.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/train.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/train_slurm.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/validate.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 70000 \
+    --validate

models/cmp/experiments/rep_learning/alexnet_yfcc+youtube_voc_16gpu_140k/validate_slurm.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py --config $work_path/config.yaml --launcher slurm \
+        --load-iter 70000 \
+        --validate

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/config.yaml

@@ -0,0 +1,58 @@
+model:
+    arch: CMP
+    total_iter: 70000
+    lr_steps: [40000, 60000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: alexnet_fcn_32x
+        sparse_encoder: shallownet32x
+        flow_decoder: MotionDecoderPlain
+        skip_layer: False
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 12
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 416
+    crop_size: [384, 384]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 0.00015625
+    nms_ks: 41
+    max_num_guide: 150
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/yfcc/lists/train.txt
+    val_source:
+        - data/yfcc/lists/val.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 10000
+    save_freq: 10000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: False

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/resume.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/train.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/train_slurm.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/validate.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 70000 \
+    --validate

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_16gpu_70k/validate_slurm.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py --config $work_path/config.yaml --launcher slurm \
+        --load-iter 70000 \
+        --validate

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/config.yaml

@@ -0,0 +1,58 @@
+model:
+    arch: CMP
+    total_iter: 140000
+    lr_steps: [80000, 120000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: alexnet_fcn_32x
+        sparse_encoder: shallownet32x
+        flow_decoder: MotionDecoderPlain
+        skip_layer: False
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 12
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 416
+    crop_size: [384, 384]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 0.00015625
+    nms_ks: 41
+    max_num_guide: 150
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/yfcc/lists/train.txt
+    val_source:
+        - data/yfcc/lists/val.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 10000
+    save_freq: 10000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: False

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/resume.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/train.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/train_slurm.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/validate.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 70000 \
+    --validate

models/cmp/experiments/rep_learning/alexnet_yfcc_voc_8gpu_140k/validate_slurm.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py --config $work_path/config.yaml --launcher slurm \
+        --load-iter 70000 \
+        --validate

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/config.yaml

@@ -0,0 +1,61 @@
+model:
+    arch: CMP
+    total_iter: 70000
+    lr_steps: [40000, 60000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: resnet50
+        sparse_encoder: shallownet8x
+        flow_decoder: MotionDecoderPlain
+        skip_layer: False
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 10
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 416
+    crop_size: [320, 320]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 0.00015625
+    nms_ks: 15
+    max_num_guide: -1
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/yfcc/lists/train.txt
+        - data/youtube9000/lists/train.txt
+        - data/VIP/lists/train.txt
+        - data/MPII/lists/train.txt
+    val_source:
+        - data/yfcc/lists/val.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 10000
+    save_freq: 10000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: False

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/resume.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/train.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/train_slurm.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/validate.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 70000 \
+    --validate

models/cmp/experiments/rep_learning/resnet50_yfcc+youtube+vip+mpii_lip_16gpu_70k/validate_slurm.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py --config $work_path/config.yaml --launcher slurm \
+        --load-iter 70000 \
+        --validate

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/config.yaml

@@ -0,0 +1,58 @@
+model:
+    arch: CMP
+    total_iter: 42000
+    lr_steps: [24000, 36000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: resnet50
+        sparse_encoder: shallownet8x
+        flow_decoder: MotionDecoderPlain
+        skip_layer: False
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 16
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 333
+    crop_size: [256, 256]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 0.00005632
+    nms_ks: 49
+    max_num_guide: -1
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/yfcc/lists/train.txt
+    val_source:
+        - data/yfcc/lists/val.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 10000
+    save_freq: 10000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: False

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/resume.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/train.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/train_slurm.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/validate.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 70000 \
+    --validate

models/cmp/experiments/rep_learning/resnet50_yfcc_coco_16gpu_42k/validate_slurm.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py --config $work_path/config.yaml --launcher slurm \
+        --load-iter 70000 \
+        --validate

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/config.yaml

@@ -0,0 +1,58 @@
+model:
+    arch: CMP
+    total_iter: 42000
+    lr_steps: [24000, 36000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: resnet50
+        sparse_encoder: shallownet8x
+        flow_decoder: MotionDecoderPlain
+        skip_layer: False
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 10
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 416
+    crop_size: [320, 320]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 0.00003629
+    nms_ks: 67
+    max_num_guide: -1
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/yfcc/lists/train.txt
+    val_source:
+        - data/yfcc/lists/val.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 10000
+    save_freq: 10000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: False

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/resume.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/train.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 \
+    --nnodes=2 --node_rank=$1 \
+    --master_addr="192.168.1.1" main.py \
+    --config $work_path/config.yaml --launcher pytorch

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/train_slurm.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n16 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/validate.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 70000 \
+    --validate

models/cmp/experiments/rep_learning/resnet50_yfcc_voc_16gpu_42k/validate_slurm.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py --config $work_path/config.yaml --launcher slurm \
+        --load-iter 70000 \
+        --validate

models/cmp/experiments/semiauto_annot/resnet50_vip+mpii_liteflow/checkpoints/ckpt_iter_42000.pth.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd3a385e227c29f89b5c7c6f4c89d356f6022fa7fcfc71ab1bd40e9833048dd6
+size 228465722

models/cmp/experiments/semiauto_annot/resnet50_vip+mpii_liteflow/config.yaml

@@ -0,0 +1,59 @@
+model:
+    arch: CMP
+    total_iter: 42000
+    lr_steps: [24000, 36000]
+    lr_mults: [0.1, 0.1]
+    lr: 0.1
+    optim: SGD
+    warmup_lr: []
+    warmup_steps: []
+    module:
+        arch: CMP
+        image_encoder: resnet50
+        sparse_encoder: shallownet8x
+        flow_decoder: MotionDecoderSkipLayer
+        skip_layer: True
+        img_enc_dim: 256
+        sparse_enc_dim: 16
+        output_dim: 198
+        decoder_combo: [1,2,4]
+        pretrained_image_encoder: False
+        flow_criterion: "DiscreteLoss"
+        nbins: 99
+        fmax: 50
+data:
+    workers: 2
+    batch_size: 8
+    batch_size_test: 1
+    data_mean: [123.675, 116.28, 103.53] # RGB
+    data_div: [58.395, 57.12, 57.375]
+    short_size: 416
+    crop_size: [384, 384]
+    sample_strategy: ['grid', 'watershed']
+    sample_bg_ratio: 5.74e-5
+    nms_ks: 41
+    max_num_guide: -1
+
+    flow_file_type: "jpg"
+    image_flow_aug:
+        flip: False
+    flow_aug:
+        reverse: False
+        scale: False
+        rotate: False
+    train_source:
+        - data/VIP/lists/train.txt
+        - data/MPII/lists/train.txt
+    val_source:
+        - data/VIP/lists/randval.txt
+    memcached: False
+trainer:
+    initial_val: True
+    print_freq: 100
+    val_freq: 5000
+    save_freq: 5000
+    val_iter: -1
+    val_disp_start_iter: 0
+    val_disp_end_iter: 16
+    loss_record: ['loss_flow']
+    tensorboard: True

models/cmp/experiments/semiauto_annot/resnet50_vip+mpii_liteflow/resume.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch \
+    --load-iter 10000 \
+    --resume

models/cmp/experiments/semiauto_annot/resnet50_vip+mpii_liteflow/resume_slurm.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+work_path=$(dirname $0)
+partition=$1
+GLOG_vmodule=MemcachedClient=-1 srun --mpi=pmi2 -p $partition -n8 \
+    --gres=gpu:8 --ntasks-per-node=8 \
+    python -u main.py \
+        --config $work_path/config.yaml --launcher slurm \
+        --load-iter 10000 \
+        --resume

models/cmp/experiments/semiauto_annot/resnet50_vip+mpii_liteflow/train.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+work_path=$(dirname $0)
+python -m torch.distributed.launch --nproc_per_node=8 main.py \
+    --config $work_path/config.yaml --launcher pytorch