ayh015 commited on
Commit
c705aaa
·
1 Parent(s): 271210d

This is a quick test for annotation pipeline of hico

Browse files
Files changed (1) hide show
  1. scripts/pipeline_hico_test.sh +71 -0
scripts/pipeline_hico_test.sh ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env bash
2
+ set -euo pipefail
3
+
4
+ export PYTHONPATH="${PYTHONPATH:-}:./"
5
+
6
+ DATA_PATH=../datasets/HICO-Det
7
+
8
+ LONG_MODEL_PATH=./model_weights/qwen3_30b_vl_instruct/models
9
+ REFINE_MODEL_PATH=./model_weights/qwen3_8b_vl_instruct
10
+ EXAMINE_MODEL_PATH=./model_weights/qwen3_30b_vl_instruct/models
11
+
12
+ LONG_GPU_IDS=0,1
13
+ REFINE_GPU_IDS=0,1
14
+ EXAMINE_GPU_IDS=0,1
15
+
16
+ LONG_NPROC=2
17
+ REFINE_NPROC=2
18
+ EXAMINE_NPROC=2
19
+
20
+ LONG_OUT_DIR=outputs/pipeline/long
21
+ REFINE_OUT_DIR=outputs/pipeline/refine
22
+ EXAMINE_OUT_DIR=outputs/pipeline/examine
23
+
24
+ MERGED_LONG_JSON=outputs/pipeline/merged_long.json
25
+ MERGED_REFINE_JSON=outputs/pipeline/merged_refine.json
26
+ MERGED_EXAMINE_JSON=outputs/pipeline/merged_examine.json
27
+
28
+ mkdir -p "${LONG_OUT_DIR}" "${REFINE_OUT_DIR}" "${EXAMINE_OUT_DIR}"
29
+
30
+ CUDA_VISIBLE_DEVICES=${LONG_GPU_IDS} OMP_NUM_THREADS=1 torchrun --nnodes=1 --nproc_per_node=${LONG_NPROC} --master_port=25011 \
31
+ tools/annotate_hico.py \
32
+ --model-path "${LONG_MODEL_PATH}" \
33
+ --data-path "${DATA_PATH}" \
34
+ --output-dir "${LONG_OUT_DIR}" \
35
+ --max-samples 5
36
+
37
+ python3 tools/merge_json_outputs.py \
38
+ --input-dir "${LONG_OUT_DIR}" \
39
+ --pattern "labels_*.json" \
40
+ --output-path "${MERGED_LONG_JSON}"
41
+
42
+ CUDA_VISIBLE_DEVICES=${REFINE_GPU_IDS} OMP_NUM_THREADS=1 torchrun --nnodes=1 --nproc_per_node=${REFINE_NPROC} --master_port=25012 \
43
+ tools/refine_hico.py \
44
+ --model-path "${REFINE_MODEL_PATH}" \
45
+ --data-path "${DATA_PATH}" \
46
+ --annotation-path "${MERGED_LONG_JSON}" \
47
+ --output-dir "${REFINE_OUT_DIR}" \
48
+ --max-samples 5
49
+
50
+ python3 tools/merge_json_outputs.py \
51
+ --input-dir "${REFINE_OUT_DIR}" \
52
+ --pattern "refine_labels_*.json" \
53
+ --output-path "${MERGED_REFINE_JSON}"
54
+
55
+ CUDA_VISIBLE_DEVICES=${EXAMINE_GPU_IDS} OMP_NUM_THREADS=1 torchrun --nnodes=1 --nproc_per_node=${EXAMINE_NPROC} --master_port=25013 \
56
+ tools/examine_hico.py \
57
+ --model-path "${EXAMINE_MODEL_PATH}" \
58
+ --data-path "${DATA_PATH}" \
59
+ --annotation-path "${MERGED_REFINE_JSON}" \
60
+ --output-dir "${EXAMINE_OUT_DIR}" \
61
+ --max-samples 5
62
+
63
+ python3 tools/merge_json_outputs.py \
64
+ --input-dir "${EXAMINE_OUT_DIR}" \
65
+ --pattern "examiner_labels_*.json" \
66
+ --output-path "${MERGED_EXAMINE_JSON}"
67
+
68
+ echo "Pipeline complete."
69
+ echo "Long descriptions: ${MERGED_LONG_JSON}"
70
+ echo "Refined descriptions: ${MERGED_REFINE_JSON}"
71
+ echo "Examiner results: ${MERGED_EXAMINE_JSON}"