LightFC-X: Lightweight Convolutional Tracker for RGB-X Tracking

June 13, 2025 · View on GitHub

Notice:

Here rgbter-light is the initial name of our tracker, and rgbt-light is the initial name of our project

Environment Installation

prepare your environment as TBSI.

Notice: Our use pytorch version is 1.13.0

Project Paths Setup

You can also modify paths by editing these two files

lib/test/evaluation/local.py  # paths about testing

env_num is used to distinguish different development devices, we recommend setting env_num to only 0 here

			RGBD	RGBD	RGBD	RGBT	RGBT	RGBT	RGBE	RGBE
	Params	FLOPs	F	Fr	Re	PR	NPR	SR	PR	SR
LightFC-X-xxs (MobileFormer)	3.83M	0.79G	0.408	0.407	0.410	54.8	51.5	44.7	64.4	49.8
LightFC-X-xs (MobileViT)	4.19M	3.03G	0.469	0.472	0.465	60.9	57.1	48.4	65.5	51.0
LightFC-X	6.68M	5.40G	0.500	0.537	0.468	63.8	59.9	50.1	68.7	52.9

RGBD is DepTrack. RGBT is LasHeR. RGBE is VisEvent.

Download LightFC-X-xs Model: https://drive.google.com/drive/folders/1EKxYogu-3WmTMTBASR6BtU2RQN90WhQp?usp=sharing

Download LightFC-X-xxs Model: https://drive.google.com/drive/folders/1xevE9iynAczPhkDs-VzqNgcGhbWEVQq5?usp=sharing

Download LightFC-X Model and results: https://drive.google.com/file/d/1rmhq42s_2ZaWZuuyihuJCUMBHM7anStG/view?usp=drive_link

Data Preparation

Put LasHeR, RGBT234, VisEvent datasets' files into

lib/test/evaluation/local.py  # paths about testing

Put DepthTrack, VOT22RGBD datasets into

Depthtrack_workspace/sequence
VOT22RGBD_workspace/sequences

Training

Training method is same as OSTrack-like Library.

python tracking/train.py --script rgbter_light --config baselinev2_ep45_maevitt --save_dir ./output --mode multiple --env_num 3 --nproc_per_node 2 --use_wandb 0

Test

Our checkpoints are released in

${PROJECT_ROOT}
  -- output
      -- checkpoints
          -- lightfcx
              |-- RGBD_baseline_N3_ep45
              |-- RGBE_baseline_ep45
              |-- RGBS_baseline_ep30
              |-- RGBT_baseline_ep45
              |-- RGBT_baseline_update_ep15

Using the following cmds for eval our trackers, you have to make env_num match the env_num in lib/test/evaluation/local.py

RGB-T trackers

# S
python tracking/test.py --script lightfcx --config RGBT_baseline_ep45  --dataset lasher --debug 0 --threads 4 --num_gpus 2 --env_num 0
python tracking/test.py --script lightfcx --config RGBT_baseline_ep45  --dataset rgbt234 --debug 0 --threads 4 --num_gpus 2 --env_num 0

# ST
python tracking/test.py --script lightfcx --config RGBT_baseline_update_ep15  --dataset lasher --debug 0 --threads 4 --num_gpus 2 --env_num 0
python tracking/test.py --script lightfcx --config RGBT_baseline_update_ep15  --dataset rgbt234 --debug 0 --threads 4 --num_gpus 2 --env_num 0

RGB-E trackers

python tracking/test_rgbe.py --script_name rgbt_light --yaml_name RGBE_baseline_ep45

RGB-D trackers

DepthTrack

cd Depthtrack_workspace
vot evaluate --workspace ./ rgbter_light
vot analysis --nocache --name rgbter_light

VOT22RGBD

cd VOT22RGBD_workspace
vot evaluate --workspace ./ rgbter_light
vot analysis --nocache --name rgbter_light

RGB-S trackers

python tracking/test_rgbs.py

Evaluation Profile Model

Evaluate RGB-X parameters and Macs:

python tracking/profile_model.py

Evaluation Results

RGB-T and RGB-E

Our RGB-T and RGB-E raw results are released in

${PROJECT_ROOT}
  -- output
      -- test
          -- tracking_results
              -- lightfcx
                  |-- RGBE_baseline_ep45
                  |-- RGBE_baseline_update_ep15
                  |-- RGBT_baseline_ep45
                  |-- RGBT_baseline_update_ep15

Evaluate RGB-T results

python tracking/analysis_results.py --tracker_name lightfcx --tracker_param RGBT_baseline_ep45 --dataset lasher --env_num 2 
python tracking/analysis_results.py --tracker_name lightfcx --tracker_param RGBT_baseline_update_ep15 --dataset lasher --env_num 2

Evaluate RGB-E results

python tracking/analysis_results.py --tracker_name lightfcx --tracker_param RGBE_baseline_ep45 --dataset lasher --env_num 2 
python tracking/analysis_results.py --tracker_name lightfcx --tracker_param RGBE_baseline_update_ep15 --dataset lasher --env_num 2

RGB-D

Our RGB-D raw results are released in

Analysis of DepthTrack

Depthtrack_workspace/analysis_RGBD_baseline_ep45
Depthtrack_workspace/analysis_RGBD_baseline_N3_ep45
Depthtrack_workspace/analysis_RGBD_baseline_N3_update_ep15

Analysis of VOT22RGBD

VOT22RGBD_workspace/analysis_RGBD_baseline_ep45
VOT22RGBD_workspace/analysis_RGBD_baseline_N3_ep45
VOT22RGBD_workspace/analysis_RGBD_baseline_N3_update_ep15

RGB-S

Our RGB-D raw results are released in

${PROJECT_ROOT}
  -- output
      -- RGBPS

Evaluate RGB-S results

python tracking/analysis_rgbs.py