Regarding the issue of PSNR indicator first rising and then falling

[junhaojia]

May I ask you a few questions about video super-resolution? @xinntao
I'm using Basic VSR++, and I suspect there is a problem with my configuration file.

train_BasicVSRPP_REDS_test.yml

GENERATE TIME: Wed Sep 10 15:33:31 2025

CMD:

/public/home/jsj_sjfx_jjh/BasicSR-master/basicsr/train.py -opt /public/home/jsj_sjfx_jjh/BasicSR-master/options/train/BasicVSRPP/train_BasicVSRPP_REDS_test.yml

general settings

name: train_BasicVSRPP_REDS
model_type: VideoRecurrentModel
scale: 4
num_gpu: 1 # official: 8 GPUs
manual_seed: 0

dataset and data loader settings

datasets:
train:
name: REDS
type: REDSRecurrentDataset
dataroot_gt: /public/home/jsj_sjfx_jjh/data/data_JILIN189/train/GT/
dataroot_lq: /public/home/jsj_sjfx_jjh/data/data_JILIN189/train/LR4x/
meta_info_file: "/public/home/jsj_sjfx_jjh/data/data_JILIN189/meta_info_REDS_GT.txt"
val_partition: REDS4 # set to 'official' when use the official validation partition
test_mode: False
io_backend:
type: disk

num_frame: 15
gt_size: 256
interval_list: [1]
random_reverse: false
use_hflip: true
use_rot: true

# data loader
num_worker_per_gpu: 0
batch_size_per_gpu: 1
dataset_enlarge_ratio: 200
prefetch_mode: ~

val:
name: REDS4
type: VideoRecurrentTestDataset
dataroot_gt: /public/home/jsj_sjfx_jjh/data/data_JILIN189/eval/GT/
dataroot_lq: /public/home/jsj_sjfx_jjh/data/data_JILIN189/eval/LR4x/

cache_data: true
io_backend:
  type: disk

num_frame: -1  # not needed

network structures

network_g:
type: BasicVSRPlusPlus
mid_channels: 64
num_blocks: 7
is_low_res_input: true
spynet_path: experiments/pretrained_models/spynet_sintel_final-3d2a1287.pth

path

path:
pretrain_network_g: ~
strict_load_g: true
resume_state: ~

training settings

train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 2e-4
weight_decay: 0
betas: [0.9, 0.99]
capturable: True

scheduler:
type: CosineAnnealingRestartLR
periods: [200000]
restart_weights: [1]
eta_min: !!float 1e-7

total_iter: 200000
warmup_iter: -1 # no warm up
fix_flow: 5000
flow_lr_mul: 0.125

losses

pixel_opt:
type: CharbonnierLoss
loss_weight: 1.0
reduction: mean

validation settings

val:
val_freq: 5000
save_img: False

metrics:
psnr: # metric name, can be arbitrary
type: calculate_psnr
crop_border: 0
test_y_channel: false
better: higher # the higher, the better. Default: higher

niqe:
  type: calculate_niqe
  crop_border: 0
  better: lower  # the lower, the better

logging settings

logger:
print_freq: 100
save_checkpoint_freq: 10000
use_tb_logger: true
wandb:
project: ~
resume_id: ~

dist training settings

dist_params:
backend: nccl
port: 29500

find_unused_parameters: true