crosstyan/OpenGait
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add Gait3D support

Browse Source
This commit is contained in:
darkliang
2022-12-05 21:47:24 +08:00
parent f79997bf8d
commit 77abcc2a10
11 changed files with 5525 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
configs/baseline/baseline_Gait3D.yaml
+104
View File
@@ -0,0 +1,104 @@
data_cfg:
dataset_name: Gait3D
dataset_root: your_path
dataset_partition: datasets/Gait3D/Gait3D.json
data_in_use: [true, false]
num_workers: 1
remove_no_gallery: false # Remove probe if no gallery for it
test_dataset_name: Gait3D
evaluator_cfg:
enable_float16: true
restore_ckpt_strict: true
restore_hint: 60000
save_name: Baseline
eval_func: evaluate_Gait3D
sampler:
batch_shuffle: false
batch_size: 16
sample_type: all_ordered # all indicates whole sequence used to test, while ordered means input sequence by its natural order; Other options: fixed_unordered
frames_all_limit: 720 # limit the number of sampled frames to prevent out of memory
metric: euc # cos euc
transform:
- type: BaseSilCuttingTransform
loss_cfg:
- loss_term_weight: 1.0
margin: 0.2
type: TripletLoss
log_prefix: triplet
- loss_term_weight: 0.1
scale: 16
type: CrossEntropyLoss
log_prefix: softmax
log_accuracy: true
model_cfg:
model: Baseline
backbone_cfg:
in_channels: 1
layers_cfg: # Layers configuration for automatically model construction
- BC-64
- BC-64
- M
- BC-128
- BC-128
- M
- BC-256
- BC-256
# - M
# - BC-512
# - BC-512
type: Plain
SeparateFCs:
in_channels: 256
out_channels: 256
parts_num: 31
SeparateBNNecks:
class_num: 3000
in_channels: 256
parts_num: 31
bin_num:
- 16
- 8
- 4
- 2
- 1
optimizer_cfg:
lr: 0.1
momentum: 0.9
solver: SGD
weight_decay: 0.0005
scheduler_cfg:
gamma: 0.1
milestones: # Learning Rate Reduction at each milestones
- 20000
- 40000
- 50000
scheduler: MultiStepLR
trainer_cfg:
enable_float16: true # half_percesion float for memory reduction and speedup
fix_BN: false
log_iter: 100
with_test: 10000
restore_ckpt_strict: true
restore_hint: 0
save_iter: 10000
save_name: Baseline
sync_BN: true
total_iter: 60000
sampler:
batch_shuffle: true
batch_size:
- 32 # TripletSampler, batch_size[0] indicates Number of Identity
- 4 # batch_size[1] indicates Samples sequqnce for each Identity
frames_num_fixed: 30 # fixed frames number for training
frames_num_max: 50 # max frames number for unfixed training
frames_num_min: 25 # min frames number for unfixed traing
sample_type: fixed_unordered # fixed control input frames number, unordered for controlling order of input tensor; Other options: unfixed_ordered or all_ordered
type: TripletSampler
transform:
- type: BaseSilCuttingTransform
configs/smplgait/smplgait.yaml
+108
View File
@@ -0,0 +1,108 @@
data_cfg:
dataset_name: Gait3D
dataset_root: your_path
dataset_partition: datasets/Gait3D/Gait3D.json
data_in_use: [true, true]
num_workers: 1
remove_no_gallery: false # Remove probe if no gallery for it
test_dataset_name: Gait3D
evaluator_cfg:
enable_float16: true
restore_ckpt_strict: true
restore_hint: 60000
save_name: SMPLGait
eval_func: evaluate_Gait3D
sampler:
batch_shuffle: false
batch_size: 16
sample_type: all_ordered # all indicates whole sequence used to test, while ordered means input sequence by its natural order; Other options: fixed_unordered
frames_all_limit: 720 # limit the number of sampled frames to prevent out of memory
metric: euc # cos euc
transform:
- type: BaseSilCuttingTransform
img_w: 64
- type: NoOperation
loss_cfg:
- loss_term_weight: 1.0
margin: 0.2
type: TripletLoss
log_prefix: triplet
- loss_term_weight: 0.1
scale: 16
type: CrossEntropyLoss
log_prefix: softmax
log_accuracy: true
model_cfg:
model: SMPLGait
backbone_cfg:
in_channels: 1
layers_cfg: # Layers configuration for automatically model construction
- BC-64
- BC-64
- M
- BC-128
- BC-128
- M
- BC-256
- BC-256
# - M
# - BC-512
# - BC-512
type: Plain
SeparateFCs:
in_channels: 256
out_channels: 256
parts_num: 31
SeparateBNNecks:
class_num: 3000
in_channels: 256
parts_num: 31
bin_num:
- 16
- 8
- 4
- 2
- 1
optimizer_cfg:
lr: 0.1
momentum: 0.9
solver: SGD
weight_decay: 0.0005
scheduler_cfg:
gamma: 0.1
milestones: # Learning Rate Reduction at each milestones
- 20000
- 40000
- 50000
scheduler: MultiStepLR
trainer_cfg:
enable_float16: true # half_percesion float for memory reduction and speedup
fix_BN: false
log_iter: 100
with_test: 10000
restore_ckpt_strict: true
restore_hint: 0
save_iter: 10000
save_name: SMPLGait
sync_BN: true
total_iter: 60000
sampler:
batch_shuffle: true
batch_size:
- 32 # TripletSampler, batch_size[0] indicates Number of Identity
- 4 # batch_size[1] indicates Samples sequqnce for each Identity
frames_num_fixed: 30 # fixed frames number for training
frames_num_max: 50 # max frames number for unfixed training
frames_num_min: 25 # min frames number for unfixed traing
sample_type: fixed_unordered # fixed control input frames number, unordered for controlling order of input tensor; Other options: unfixed_ordered or all_ordered
type: TripletSampler
transform:
- type: BaseSilCuttingTransform
img_w: 64
- type: NoOperation
datasets/CASIA-B/README.md
+1 -1
View File
@@ -13,7 +13,7 @@ Download URL: http://www.cbsr.ia.ac.cn/GaitDatasetB-silh.zip
......
......
```
- Run `python misc/pretreatment.py --input_path CASIA-B --output_path CASIA-B-pkl`
- Run `python datasets/pretreatment.py --input_path CASIA-B --output_path CASIA-B-pkl`
- Processed
```
CASIA-B-pkl
datasets/Gait3D/Gait3D.json
+5008
View File
File diff suppressed because it is too large Load Diff
datasets/Gait3D/README.md
+33
View File
@@ -0,0 +1,33 @@
# Gait3D
This is the pre-processing instructions for the Gait3D dataset. The original dataset can be found [here](https://gait3d.github.io/). The original dataset is not publicly available. You need to request access to the dataset in order to download it. This README explains how to extract the original dataset and convert it to a format suitable for OpenGait.
## Data Preparation
https://github.com/Gait3D/Gait3D-Benchmark#data-preparation
## Data Pretreatment
```python
python datasets/pretreatment.py --input_path 'Gait3D/2D_Silhouettes' --output_path 'Gait3D-sils-64-64-pkl'
python datasets/pretreatment_smpl.py --input_path 'Gait3D/3D_SMPLs' --output_path 'Gait3D-smpls-pkl'
(optional) python datasets/pretreatment.py --input_path 'Gait3D/2D_Silhouettes' --img_size 128 --output_path 'Gait3D-sils-128-128-pkl'
python datasets/Gait3D/merge_two_modality.py --sils_path 'Gait3D-sils-64-64-pkl' --smpls_path 'Gait3D-smpls-pkl' --output_path 'Gait3D-merged-pkl' --link 'hard'
```
## Train
### Baseline model:
`CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs ./configs/baseline/baseline_Gait3D.yaml --phase train`
### SMPLGait model:
`CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs ./configs/smplgait/smplgait.yaml --phase train`
## Citation
If you use this dataset in your research, please cite the following paper:
```
@inproceedings{zheng2022gait3d,
title={Gait Recognition in the Wild with Dense 3D Representations and A Benchmark},
author={Jinkai Zheng, Xinchen Liu, Wu Liu, Lingxiao He, Chenggang Yan, Tao Mei},
booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2022}
}
```
## Acknowledgements
This dataset was collected by the [Zheng at. al.](https://gait3d.github.io/). The pre-processing instructions are based on (https://github.com/Gait3D/Gait3D-Benchmark).
datasets/Gait3D/merge_two_modality.py
+46
View File
@@ -0,0 +1,46 @@
import os
import argparse
from pathlib import Path
import shutil
def merge(sils_path, smpls_path, output_path, link):
if link == 'hard':
link_method = os.link
elif link == 'soft':
link_method = os.symlink
else:
link_method = shutil.copyfile
for _id in os.listdir(sils_path):
id_path = os.path.join(sils_path, _id)
for _type in os.listdir(id_path):
type_path = os.path.join(id_path, _type)
for _view in os.listdir(type_path):
view_path = os.path.join(type_path, _view)
for _seq in os.listdir(view_path):
sils_seq_path = os.path.join(view_path, _seq)
smpls_seq_path = os.path.join(
smpls_path, _id, _type, _view, _seq)
output_seq_path = os.path.join(output_path, _id, _type, _view)
os.makedirs(output_seq_path, exist_ok=True)
link_method(sils_seq_path, os.path.join(
output_seq_path, "sils-"+_seq))
link_method(smpls_seq_path, os.path.join(
output_seq_path, "smpls-"+_seq))
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Gait3D dataset mergence.')
parser.add_argument('--sils_path', default='', type=str,
help='Root path of raw silhs dataset.')
parser.add_argument('--smpls_path', default='', type=str,
help='Root path of raw smpls dataset.')
parser.add_argument('-o', '--output_path', default='',
type=str, help='Output path of pickled dataset.')
parser.add_argument('-l', '--link', default='hard', type=str,
choices=['hard', 'soft', 'copy'], help='Link type of output data.')
args = parser.parse_args()
merge(sils_path=Path(args.sils_path), smpls_path=Path(
args.smpls_path), output_path=Path(args.output_path), link=args.link)
datasets/HID/README.md
+4 -4
View File
@@ -6,19 +6,19 @@ This is the official support for competition of [Human Identification at a Dista
Download the raw dataset from the [official link](http://hid2022.iapr-tc4.org/). You will get three compressed files, i.e. `train.tar`, `HID2022_test_gallery.zip` and `HID2022_test_probe.zip`.
After unpacking these three files, run this command:
```shell
python misc/HID/pretreatment_HID.py --input_train_path="train" --input_gallery_path="HID2022_test_gallery" --input_probe_path="HID2022_test_probe" --output_path="HID-128-pkl"
python datasets/HID/pretreatment_HID.py --input_train_path="train" --input_gallery_path="HID2022_test_gallery" --input_probe_path="HID2022_test_probe" --output_path="HID-128-pkl"
```
## Train the dataset
Modify the `dataset_root` in `./misc/HID/baseline_hid.yaml`, and then run this command:
Modify the `dataset_root` in `configs/baseline/baseline_hid.yaml`, and then run this command:
```shell
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs ./misc/HID/baseline_hid.yaml --phase train
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs configs/baseline/baseline_hid.yaml --phase train
```
You can also download the [trained model](https://github.com/ShiqiYu/OpenGait/releases/download/v1.1/pretrained_hid_model.zip) and place it in `output` after unzipping.
## Get the submission file
```shell
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs ./misc/HID/baseline_hid.yaml --phase test
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs configs/baseline/baseline_hid.yaml --phase test
```
The result will be generated in your working directory.
docs/1.model_zoo.md
+9 -7
View File
@@ -39,16 +39,18 @@
| [GaitPart(CVPR2020)](http://home.ustc.edu.cn/~saihui/papers/cvpr2020_gaitpart.pdf) | 47.6 | [gaitpart.yaml](../configs/gaitpart/gaitpart_GREW.yaml) | 64x44 | - | - |
| [GaitGL(ICCV2021)](https://openaccess.thecvf.com/content/ICCV2021/papers/Lin_Gait_Recognition_via_Effective_Global-Local_Feature_Representation_and_Local_Temporal_ICCV_2021_paper.pdf) | 41.5 | [gaitgl.yaml](../configs/gaitgl/gaitgl_GREW.yaml) | 64x44 | - | - |
| [GaitGL(BNNeck)(ICCV2021)](https://openaccess.thecvf.com/content/ICCV2021/papers/Lin_Gait_Recognition_via_Effective_Global-Local_Feature_Representation_and_Local_Temporal_ICCV_2021_paper.pdf) | 51.7 | [gaitgl.yaml](../configs/gaitgl/gaitgl_GREW_BNNeck.yaml) | 64x44 | - | - |
| [GaitBase](https://arxiv.org/abs/2211.06597) | 60.1 | -| 64x44 | - | - |
| [GaitBase](https://arxiv.org/abs/2211.06597) | 60.1 | - | 64x44 | - | - |
## [Gait3D](https://github.com/Gait3D/Gait3D-Benchmark)
| Model | `Rank@1` | Configuration | Input Size | Inference Time | Model Size |
| :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :------: | :-----------: | :--------: | :------------: | :--------: |
| [GaitSet(AAAI2019)](https://arxiv.org/pdf/1811.06186.pdf) | 36.7 | - | 64x44 | - | - |
| [GaitPart(CVPR2020)](http://home.ustc.edu.cn/~saihui/papers/cvpr2020_gaitpart.pdf) | 28.2 | - | 64x44 | - | - |
| [GaitGL(ICCV2021)](https://openaccess.thecvf.com/content/ICCV2021/papers/Lin_Gait_Recognition_via_Effective_Global-Local_Feature_Representation_and_Local_Temporal_ICCV_2021_paper.pdf) | 29.7 | - | 64x44 | - | - |
| [GaitBase](https://arxiv.org/abs/2211.06597) | 65.6 | - | 64x44 | - | - |
| Model | `Rank@1` | Configuration | Input Size | Input modality |
| :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :------: | :-----------: | :--------: | :--------:|
| Baseline | 46.9 | [baseling_Gait3D.yaml](../configs/baseline/baseline_Gait3D.yaml) | 64x44 | Silhouette |
| [GaitSet(AAAI2019)](https://arxiv.org/pdf/1811.06186.pdf) | - | - | 64x44 | Silhouette |
| [GaitPart(CVPR2020)](http://home.ustc.edu.cn/~saihui/papers/cvpr2020_gaitpart.pdf) | - | - | 64x44 | Silhouette |
| [GaitGL(ICCV2021)](https://openaccess.thecvf.com/content/ICCV2021/papers/Lin_Gait_Recognition_via_Effective_Global-Local_Feature_Representation_and_Local_Temporal_ICCV_2021_paper.pdf) | - | - | 64x44 | Silhouette |
| [SMPLGait(CVPR 2022)](https://gait3d.github.io/) | 48.3(46.3) | [smplgait.yaml](../configs/smplgait/smplgait.yaml) | 64x44 | Silhouette + RGB|
| [GaitBase](https://arxiv.org/abs/2211.06597) | 65.6 | - | 64x44 | Silhouette |
------------------------------------------
opengait/evaluation/evaluator.py
+41 -1
View File
@@ -3,7 +3,7 @@ from time import strftime, localtime
import numpy as np
from utils import get_msg_mgr, mkdir
from .metric import mean_iou, cuda_dist, compute_ACC_mAP
from .metric import mean_iou, cuda_dist, compute_ACC_mAP, evaluate_rank
from .re_rank import re_ranking
@@ -225,3 +225,43 @@ def evaluate_segmentation(data, dataset):
miou = mean_iou(pred, labels)
get_msg_mgr().log_info('mIOU: %.3f' % (miou.mean()))
return {"scalar/test_accuracy/mIOU": miou}
def evaluate_Gait3D(data, conf, metric='euc'):
msg_mgr = get_msg_mgr()
features, labels, cams, time_seqs = data['embeddings'], data['labels'], data['types'], data['views']
import json
probe_sets = json.load(
open('./datasets/Gait3D/Gait3D.json', 'rb'))['PROBE_SET']
probe_mask = []
for id, ty, sq in zip(labels, cams, time_seqs):
if '-'.join([id, ty, sq]) in probe_sets:
probe_mask.append(True)
else:
probe_mask.append(False)
probe_mask = np.array(probe_mask)
# probe_features = features[:probe_num]
probe_features = features[probe_mask]
# gallery_features = features[probe_num:]
gallery_features = features[~probe_mask]
# probe_lbls = np.asarray(labels[:probe_num])
# gallery_lbls = np.asarray(labels[probe_num:])
probe_lbls = np.asarray(labels)[probe_mask]
gallery_lbls = np.asarray(labels)[~probe_mask]
results = {}
msg_mgr.log_info(f"The test metric you choose is {metric}.")
dist = cuda_dist(probe_features, gallery_features, metric).cpu().numpy()
cmc, all_AP, all_INP = evaluate_rank(dist, probe_lbls, gallery_lbls)
mAP = np.mean(all_AP)
mINP = np.mean(all_INP)
for r in [1, 5, 10]:
results['scalar/test_accuracy/Rank-{}'.format(r)] = cmc[r - 1] * 100
results['scalar/test_accuracy/mAP'] = mAP * 100
results['scalar/test_accuracy/mINP'] = mINP * 100
# print_csv_format(dataset_name, results)
msg_mgr.log_info(results)
return results
opengait/evaluation/metric.py
+70
View File
@@ -86,3 +86,73 @@ def compute_ACC_mAP(distmat, q_pids, g_pids, q_views=None, g_views=None, rank=1)
mAP = np.mean(all_AP)
return ACC, mAP
def evaluate_rank(distmat, p_lbls, g_lbls, max_rank=50):
'''
Copy from https://github.com/Gait3D/Gait3D-Benchmark/blob/72beab994c137b902d826f4b9f9e95b107bebd78/lib/utils/rank.py#L12-L63
'''
num_p, num_g = distmat.shape
if num_g < max_rank:
max_rank = num_g
print('Note: number of gallery samples is quite small, got {}'.format(num_g))
indices = np.argsort(distmat, axis=1)
matches = (g_lbls[indices] == p_lbls[:, np.newaxis]).astype(np.int32)
# compute cmc curve for each probe
all_cmc = []
all_AP = []
all_INP = []
num_valid_p = 0. # number of valid probe
for p_idx in range(num_p):
# compute cmc curve
# binary vector, positions with value 1 are correct matches
raw_cmc = matches[p_idx]
if not np.any(raw_cmc):
# this condition is true when probe identity does not appear in gallery
continue
cmc = raw_cmc.cumsum()
pos_idx = np.where(raw_cmc == 1) # 返回坐标，此处raw_cmc为一维矩阵，所以返回相当于index
max_pos_idx = np.max(pos_idx)
inp = cmc[max_pos_idx] / (max_pos_idx + 1.0)
all_INP.append(inp)
cmc[cmc > 1] = 1
all_cmc.append(cmc[:max_rank])
num_valid_p += 1.
# compute average precision
# reference: https://en.wikipedia.org/wiki/Evaluation_measures_(information_retrieval)#Average_precision
num_rel = raw_cmc.sum()
pos_idx = np.where(raw_cmc == 1) # 返回坐标，此处raw_cmc为一维矩阵，所以返回相当于index
max_pos_idx = np.max(pos_idx)
inp = cmc[max_pos_idx] / (max_pos_idx + 1.0)
all_INP.append(inp)
cmc[cmc > 1] = 1
all_cmc.append(cmc[:max_rank])
num_valid_p += 1.
# compute average precision
# reference: https://en.wikipedia.org/wiki/Evaluation_measures_(information_retrieval)#Average_precision
num_rel = raw_cmc.sum()
tmp_cmc = raw_cmc.cumsum()
tmp_cmc = [x / (i + 1.) for i, x in enumerate(tmp_cmc)]
tmp_cmc = np.asarray(tmp_cmc) * raw_cmc
AP = tmp_cmc.sum() / num_rel
all_AP.append(AP)
assert num_valid_p > 0, 'Error: all probe identities do not appear in gallery'
all_cmc = np.asarray(all_cmc).astype(np.float32)
all_cmc = all_cmc.sum(0) / num_valid_p
return all_cmc, all_AP, all_INP
opengait/modeling/models/smplgait.py
+101
View File
@@ -0,0 +1,101 @@
'''
Modifed from https://github.com/Gait3D/Gait3D-Benchmark/blob/72beab994c137b902d826f4b9f9e95b107bebd78/lib/modeling/models/smplgait.py
'''
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
from ..base_model import BaseModel
from ..modules import SetBlockWrapper, HorizontalPoolingPyramid, PackSequenceWrapper, SeparateFCs, SeparateBNNecks
class SMPLGait(BaseModel):
def __init__(self, cfgs, is_training):
super().__init__(cfgs, is_training)
def build_network(self, model_cfg):
# Baseline
self.Backbone = self.get_backbone(model_cfg['backbone_cfg'])
self.Backbone = SetBlockWrapper(self.Backbone)
self.FCs = SeparateFCs(**model_cfg['SeparateFCs'])
self.BNNecks = SeparateBNNecks(**model_cfg['SeparateBNNecks'])
self.TP = PackSequenceWrapper(torch.max)
self.HPP = HorizontalPoolingPyramid(bin_num=model_cfg['bin_num'])
# for SMPL
self.fc1 = nn.Linear(85, 128)
self.fc2 = nn.Linear(128, 256)
self.fc3 = nn.Linear(256, 256)
self.bn1 = nn.BatchNorm1d(128)
self.bn2 = nn.BatchNorm1d(256)
self.bn3 = nn.BatchNorm1d(256)
self.dropout2 = nn.Dropout(p=0.2)
self.dropout3 = nn.Dropout(p=0.2)
def forward(self, inputs):
ipts, labs, _, _, seqL = inputs
sils = ipts[0] # [n, s, h, w]
smpls = ipts[1] # [n, s, d]
# extract SMPL features
n, s, d = smpls.size()
sps = smpls.view(-1, d)
del smpls
sps = F.relu(self.bn1(self.fc1(sps)))
sps = F.relu(self.bn2(self.dropout2(self.fc2(sps)))) # (B, 256)
sps = F.relu(self.bn3(self.dropout3(self.fc3(sps)))) # (B, 256)
sps = sps.reshape(n, 1, s, 16, 16)
iden = Variable(torch.eye(16)).unsqueeze(
0).repeat(n, 1, s, 1, 1) # [n, 1, s, 16, 16]
if sps.is_cuda:
iden = iden.cuda()
sps_trans = sps + iden # [n, 1, s, 16, 16]
if len(sils.size()) == 4:
sils = sils.unsqueeze(1)
del ipts
outs = self.Backbone(sils) # [n, c, s, h, w]
outs_n, outs_c, outs_s, outs_h, outs_w = outs.size()
zero_tensor = Variable(torch.zeros(
(outs_n, outs_c, outs_s, outs_h, outs_h-outs_w)))
if outs.is_cuda:
zero_tensor = zero_tensor.cuda()
# [n, s, c, h, h] [n, s, c, 16, 16]
outs = torch.cat([outs, zero_tensor], -1)
outs = outs.reshape(outs_n*outs_c*outs_s, outs_h,
outs_h) # [n*c*s, 16, 16]
sps = sps_trans.repeat(1, outs_c, 1, 1, 1).reshape(
outs_n * outs_c * outs_s, 16, 16)
outs_trans = torch.bmm(outs, sps)
outs_trans = outs_trans.reshape(outs_n, outs_c, outs_s, outs_h, outs_h)
# Temporal Pooling, TP
outs_trans = self.TP(outs_trans, seqL, options={"dim": 2})[
0] # [n, c, h, w]
# Horizontal Pooling Matching, HPM
feat = self.HPP(outs_trans) # [n, c, p]
embed_1 = self.FCs(feat) # [n, c, p]
embed_2, logits = self.BNNecks(embed_1) # [n, c, p]
n, _, s, h, w = sils.size()
retval = {
'training_feat': {
'triplet': {'embeddings': embed_1, 'labels': labs},
'softmax': {'logits': logits, 'labels': labs}
},
'visual_summary': {
'image/sils': sils.view(n*s, 1, h, w)
},
'inference_feat': {
'embeddings': embed_1
}
}
return retval