shreyaskamathkm/ftnet

Pytorch implementation of FTNet for Semantic Segmentation on SODA, SCUT Seg, and MFN Datasets

ftnet's Issues

Finetuning | --pretrain-checkpoints not recognized

I'm trying to finetune the model as specified by the scripts provided but I encounter the following error:

usage: main.py [-h] [--mode {train,test,train_test}] [--train_only TRAIN_ONLY] [--model MODEL] [--backbone BACKBONE] [--pretrained-base PRETRAINED_BASE]
               [--dilation DILATION] [--dataset {cityscapes_thermal_combine,soda,mfn,scutseg}] [--dataset-path DATASET_PATH] [--base-size BASE_SIZE]
               [--crop-size CROP_SIZE] [--workers WORKERS] [--no-of-filters NO_OF_FILTERS] [--edge-extracts EDGE_EXTRACTS] [--num-blocks NUM_BLOCKS]
               [--train-batch-size TRAIN_BATCH_SIZE] [--val-batch-size VAL_BATCH_SIZE] [--test-batch-size TEST_BATCH_SIZE]
               [--accumulate-grad-batches ACCUMULATE_GRAD_BATCHES] [--test-monitor TEST_MONITOR] [--test-monitor-path TEST_MONITOR_PATH] [--wandb-id WANDB_ID]
               [--wandb-name-ext WANDB_NAME_EXT] [--epochs EPOCHS] [--loss-weight LOSS_WEIGHT] [--optimizer {SGD,ADAM,RMSprop,AdaBound}] [--lr LR]
               [--momentum MOMENTUM] [--nesterov NESTEROV] [--weight-decay WEIGHT_DECAY] [--beta1 BETA1] [--beta2 BETA2] [--epsilon EPSILON]
               [--scheduler-type {step,multistep_90_160,poly_warmstartup,multistep_warmstartup,onecycle}] [--warmup-iters WARMUP_ITERS]
               [--warmup-factor WARMUP_FACTOR] [--warmup-method WARMUP_METHOD] [--gamma GAMMA] [--resume RESUME] [--save-dir SAVE_DIR]
               [--test-checkpoint TEST_CHECKPOINT] [--save-images SAVE_IMAGES] [--save-images-as-subplots SAVE_IMAGES_AS_SUBPLOTS] [--debug DEBUG] [--seed SEED]
               [--num-nodes NUM_NODES] [--gpus GPUS] [--distributed-backend {dp,ddp,ddp2,horovod}]
main.py: error: unrecognized arguments: --pretrain-checkpoint ./../../../Training_Paper/Lightning/ftnet_resnext50_32x4d_128_3_1/cityscapes_thermal_combine//ckpt/last.ckpt

I'm not sure why main.py is unable to recognize the argument.

no code under folders

Hi,
nice ideas in the paper.
Do you intend to share code?
Thanks in advance

Inference on single image\folder

Hi,
thanks for the great work.
I want to test some of the pre-trained models on my own data.
is there some simple way to run a pre-trained model on a folder of images? or on a single image?
thanks

Cuda Out of Memory Issue

Hello,
I am experiencing cuda out of memory issue when run 'train_and_test.bash'.

The card is Titan RTX with 24GB memory, however, the log said

RuntimeError: CUDA out of memory. Tried to allocate 7.05GiB (GPU 0; 23.65 GiB total capacity; 16.73 GiB already allocated; 5.36GiB free; 17.00 GiB reserved in total by PyTorch)

Is there anything I should change to run training ?

ModuleNotFoundError: No module named 'core.utils.download'

Description

There is no core.utils.download but it's being imported in the init.py file under core/utils.

Log

Traceback (most recent call last):
  File "./../lightning_scripts/main.py", line 19, in <module>
    from trainers import thermal_edge_trainer
  File "/home/hmohan/Programming/T-SegModels/FTNet/Codes/src/lightning_scripts/trainers/__init__.py", line 1, in <module>
    from .thermal_edge_trainer import SegmentationLightningModel as thermal_edge_trainer
  File "/home/hmohan/Programming/T-SegModels/FTNet/Codes/src/lightning_scripts/trainers/thermal_edge_trainer.py", line 9, in <module>
    from .base_trainer import BaseTrainer
  File "/home/hmohan/Programming/T-SegModels/FTNet/Codes/src/lightning_scripts/trainers/base_trainer.py", line 26, in <module>
    from core.utils.json_extension import save_to_json_pretty
  File "/home/hmohan/Programming/T-SegModels/FTNet/Codes/src/core/utils/__init__.py", line 4, in <module>
    from .download import *
ModuleNotFoundError: No module named 'core.utils.download'